Marine ship-unmanned aerial vehicle collaborative risk assessment method fusing fuzzy risk and multi-graph association

Abstract

The invention discloses a marine ship-unmanned aerial vehicle collaborative risk assessment method fusing fuzzy risk and multi-graph association, and belongs to the field of marine traffic safety management, and the method comprises the following steps: S1, carrying out the preprocessing and standardization of multi-source data, and outputting a standardized data set with a unified reference; s2, associating the navigation space adaptive grid diagram, the ship trajectory diagram, the unmanned aerial vehicle trajectory diagram and the environmental risk diagram through a hidden Markov model, and outputting an optimal association path and matching confidence; and S3, based on the optimal association path, quantifying the risk, dynamically optimizing the risk weight by adopting deep reinforcement learning, and outputting a comprehensive fuzzy risk value and a risk level. By adopting the offshore ship-unmanned aerial vehicle collaborative risk assessment method fusing the fuzzy risk and the multi-graph association, accurate assessment and dynamic management and control of the offshore ship-unmanned aerial vehicle collaborative operation risk are realized, and the problems of offshore environment fuzziness and multi-source data space dislocation are solved; and evaluation accuracy, environmental adaptability and operation safety are also considered.

Description

Technical Field

[0001] This invention relates to the field of maritime traffic safety management technology, and in particular to a maritime ship-UAV collaborative risk assessment method that integrates fuzzy risk and multi-map association. Background Technology

[0002] With the continuous expansion of global maritime activities, collaborative operations between ships and drones have become an important development direction in fields such as maritime patrol, resupply, emergency rescue, and environmental monitoring. By providing endurance support and mission coordination through ships, and enabling flexible and rapid close-range operations through drones, the collaboration between the two can significantly improve the efficiency and coverage of maritime operations. Especially in complex sea areas (such as waters with multiple islands or busy near-shore channels) or in adverse sea conditions, the collaborative model can overcome the limitations of single-ship or drone operations and reduce labor costs and operational risks.

[0003] However, the marine environment is characterized by significant dynamism and uncertainty, and the resulting "fuzzy risks" are a core challenge in risk assessment for collaborative operations. For example, sudden changes in wind and wave conditions can lead to ship stall and course deviation, and UAV lift loss and attitude fluctuations; nonlinear deviations in ship control systems and the randomness of UAV communication delays further exacerbate the fuzziness of risks. Existing technologies mostly use static empirical formulas or simplified models to quantify such risks, which cannot dynamically adapt to real-time changes in the environment and equipment status. This easily leads to overestimation of risks (resulting in excessive avoidance and low operational efficiency) or underestimation (causing collisions, equipment failures, and other safety accidents), making it difficult to accurately reflect the actual risk level of collaborative operations. Summary of the Invention

[0004] The purpose of this invention is to provide a method for collaborative risk assessment of maritime vessels and unmanned aerial vehicles that integrates fuzzy risk and multi-map association, thereby solving the aforementioned technical problems.

[0005] To achieve the above objectives, this invention provides a maritime ship-UAV collaborative risk assessment method that integrates fuzzy risk and multi-graph association, comprising the following steps: S1. Multi-source data preprocessing and standardization: Simultaneously collect raw ship AIS data, UAV flight data and environmental data to form a raw dataset, and clean, smooth, spatiotemporally interpolate and standardize the raw dataset to output a standardized dataset with a unified benchmark. S2. Multi-graph construction and cross-graph association: Based on standardized data, construct an adaptive grid map of navigation space, ship trajectory map, UAV trajectory map and environmental risk map, and associate the adaptive grid map of navigation space, ship trajectory map, UAV trajectory map and environmental risk map through a hidden Markov model, and output the optimal association path and matching confidence. S3. Fuzzy Risk Factor Quantification and Modeling: Based on the optimal correlation path, the dynamic risks of ships, drones and collaborative interactions are quantified, and deep reinforcement learning is used to dynamically optimize the risk weights, outputting a comprehensive fuzzy risk value and risk level.

[0006] Therefore, the above-mentioned maritime ship-UAV collaborative risk assessment method that integrates fuzzy risk and multi-graph association has the following beneficial effects: 1. Improve the availability of multi-source data and lay the foundation for assessment: By cleaning, smoothing, spatiotemporal bilinear interpolation and dimensional standardization, the dimensional benchmarks of ship AIS data, UAV flight data (including sensor noise variance), ECMWF meteorological data and CFD performance data are established to eliminate problems such as data format differences, spatiotemporal misalignment and noise interference, and ensure the consistency and availability of multi-source data, providing high-quality data support for subsequent multi-map construction and risk quantification; 2. Ensure the accuracy of spatial association among multiple maps and solve the spatial alignment problem: Construct a navigation spatial grid map based on an adaptive quadtree, combine it with ship / UAV trajectory maps and environmental risk maps, and achieve cross-map association of the four maps through a Hidden Markov Model (HMM); use the Viterbi algorithm to solve for the optimal association path, and the matching confidence level must be ≥0.8 to avoid the spatial misalignment problem of traditional single map matching, and ensure accurate spatial alignment between ship-UAV trajectories and environmental risks, providing a reliable spatial benchmark for risk assessment; 3. Achieve precise quantification of fuzzy risks, covering all collaborative scenarios: comprehensively covering ship dynamic risks (CFD-corrected stall risk, heading deviation risk), UAV dynamic risks (lift loss risk, attitude deviation risk), and collaborative interaction risks (collision risk, communication risk, mission deviation risk); by correcting environmental risk calculations through CFD coefficients and combining fuzzy membership functions to handle uncertainties in the marine environment, the problem of overestimation / underestimation of risks in traditional static empirical formulas is avoided, thereby improving the accuracy of risk quantification; 4. Enhance the dynamic adaptability of risk assessment to adapt to complex marine environments: Deep reinforcement learning (DRL) is used to dynamically optimize the aggregate weights of three types of risks. The state space includes environmental risks, ship / drone / cooperative risks. The reward function aims to minimize the risk assessment error. It can respond in real time to dynamic scenarios such as sudden changes in wind and waves and communication delay fluctuations, ensuring that the assessment results adapt to the dynamic changes in the marine environment.

[0007] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0008] Figure 1 This is a flowchart of the maritime ship-UAV collaborative risk assessment method that integrates fuzzy risk and multi-graph association according to the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are merely illustrative of the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0010] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0011] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0012] like Figure 1 As shown, the maritime ship-UAV collaborative risk assessment method integrating fuzzy risk and multi-map association includes the following steps: S1. Multi-source data preprocessing and standardization: Simultaneously collect raw ship AIS data, UAV flight data and environmental data to form a raw dataset, and clean, smooth, spatiotemporally interpolate and standardize the raw dataset to output a standardized dataset with a unified benchmark. The standardized dataset described in step S1 , These represent standardized ship AIS data, drone flight data, and environmental data, respectively; among them, , These represent the ship's longitude, latitude, altitude, speed, heading, and timestamp, respectively. , These represent the longitude, latitude, altitude, speed, attitude angle, sensor noise variance, and timestamp of the UAV, respectively; environmental data. ECMWF meteorological data were analyzed using spatiotemporal bilinear interpolation. Discrete data is converted into continuous grid data and then fused with CFD performance data. Acquisition of ECMWF meteorological data , Representing wave height, wave direction, and wave period respectively, CFD performance data. , These represent the ship stall coefficient and the UAV lift loss coefficient, respectively. Spatiotemporal bilinear interpolation includes spatial bilinear interpolation and temporal linear interpolation: ; ; In the formula, Indicates the target location Interpolation results of environmental parameters at the location; express Any parameter in; Indicates surrounding Environmental parameter values ​​for the four selected ECMWF grid points; They represent the positions around the target location. ECMWF grid points were obtained; Indicates the target time Interpolation results of environmental parameters at the location; and They represent respectively; and Representing the target time respectively Interpolation results of environmental parameters at two different time points; Fusion CFD performance data The process is as follows: First, establish the ship type and ship stall coefficient respectively. And the drone model and the drone lift loss coefficient A mapping table between them is used to determine the lift loss coefficient of the UAV based on the current ship type and UAV model. and ship stall coefficient Finally, the lift loss coefficient of the drone was calculated. and ship stall coefficient Assign the value to the corresponding ECMWF mesh.

[0013] S2. Multi-graph construction and cross-graph association: Based on standardized data, construct an adaptive grid map of navigation space, ship trajectory map, UAV trajectory map and environmental risk map, and associate the adaptive grid map of navigation space, ship trajectory map, UAV trajectory map and environmental risk map through a hidden Markov model, and output the optimal association path and matching confidence. Step S2 specifically includes the following steps: S21. Construct an adaptive grid map of the navigation space: Use an adaptive quadtree to divide the collaborative operation area, in the ship-UAV interaction area. Perform grid division to generate an adaptive grid map of navigation space. ;in, This indicates the latitude and longitude coordinates of the lower left corner of the ship-drone collaborative operation area. This indicates the latitude and longitude coordinates of the upper right corner of the ship-drone collaborative operation area; Represents grid encoding; Indicates the grid type, and ; Indicates the coordinates of the grid center, and , and Let represent the grid side lengths of the adaptive quadtree in the longitude and latitude directions, respectively. , Indicates the depth of the partition; S22. Constructing a ship trajectory map: First, the ship's trajectory points Determine the matching grid by coordinate attribution: ; In the formula, The grid to which the ship's trajectory point belongs; Then connect the consecutive trajectory points in chronological order. If two adjacent ship trajectory points and The grid encoding satisfies Then a trajectory segment is generated. ; in, They represent the first The and the first The grid code to which each ship's trajectory point belongs; They represent the first The and the first Timestamps of each ship's trajectory point; This represents the average velocity of the trajectory segment, and , and They represent the first The and the first The actual speed of each ship's trajectory point; Finally, statistics for each grid. Number of trajectory segments : ; In the formula, Indicates an indicator function; and Representing trajectory segments The starting and ending grids; Represents trajectory segment A set; Obtain ship trajectory map ; S23. Construct a drone trajectory map; First, the drone's trajectory points Match to the corresponding grid: ; In the formula, The grid to which the drone's trajectory points belong; Then connect the consecutive trajectory points in chronological order. If two adjacent trajectory points... and The grid encoding satisfies Then a trajectory segment is generated. ; in, This represents the average velocity of the trajectory segment, and , and They represent the first The and the first The actual speed of each drone trajectory point; Indicates the average flight altitude of the trajectory segment; Finally, statistics for each grid. Number of trajectory segments : ; In the formula, Indicates an indicator function; Represents trajectory segment A set; Obtain ship trajectory map ; S24. Constructing an environmental risk map: Integrating standardized environmental data with an adaptive grid map of airspace. By correlating and quantifying the environmental threat level of each grid, an environmental risk map can be constructed. S241. Calculate the comprehensive wind and wave risk : ; in, ,and ; ; ; ,and ; ; In the formula, and Indicates the collaborative weight; and These respectively represent the wind and wave risks for ships and the wind and wave risks for drones; This indicates the rate of speed loss of a ship due to wind and waves; Indicates the ship's maximum speed; Indicates the angle between the ship's heading and the wave direction; Indicates the lift loss rate of the drone; This indicates the angle between the drone's attitude angle and the wave direction; S242. Calculate No-Fly and No-Sail Risks: Set no-fly and no-sail risks for no-fly and no-sail grids. Regarding the risks of no-fly and no-navigation zones in the general aviation grid settings. ; S243, Taking into account comprehensive wind and wave risks Risks of no-fly and no-fly zones The maximum value is used as the total grid risk value. : ; S25. Cross-graph association: Connect the adaptive grid map of navigation space, ship trajectory map, UAV trajectory map and environmental risk map through hidden Markov model, and output the optimal association path with the grid as the hidden state and the trajectory point as the observation value. S251. Define a Hidden Markov Model: Define the hidden state space. Adaptive mesh map for airspace Navigation grid in the observation space; Elements in the state transition matrix , Represents a grid Transfer to grid The probability, and , Indicates the risk discount factor. Represents a grid The number of adjacent navigation grids; elements in the observation matrix , Represents a grid Generate observations The probability of; Calculate the observation-grid matching probability : ; in, ; ; ; ; In the formula, Indicates weight; This indicates the probability of a match between ship observations and the grid. This indicates the probability of a match between UAV observations and the grid. This represents the distance from the ship's trajectory point to the center of the grid. Indicates the ship's positioning error; This represents the distance from the drone's trajectory point to the center of the grid. This indicates the drone's positioning error; S252. Use the Viterbi algorithm to solve for the optimal association path and confidence of the Hidden Markov Model, and obtain the multi-graph association results. , Indicates the optimal association path. Indicates the match confidence level; S253, Determine whether the condition is satisfied. If yes, proceed to step S3; otherwise, return to step S21.

[0014] In step S21, the mesh type determination strategy is as follows: If the grid node is entirely within a no-fly or no-navigation zone: ; If the grid nodes are entirely within the navigable area: ; If grid nodes partially overlap, then mark them as grid nodes to be subdivided. For a node to be subdivided, further subdivision to child nodes is allowed if one of the following conditions is met: Ship-UAV trajectory density within nodes , Indicates the number of ship-drone trajectory segments; Indicates the grid area; This indicates setting a trajectory density threshold; Distance between the node and the boundary of the no-fly or no-navigation zone , This indicates setting a boundary distance threshold; Furthermore, the subdivided child node encoding rules are as follows: assuming the parent node is encoded as... The codes for the four child nodes are as follows: .

[0015] Step S252 specifically includes the following steps: S2521. Initialization: Initialize all navigation grids at the first moment to generate the first observation. The maximum probability, and the previous state pointer: ; ; In the formula, Indicates that the first moment is in the grid The maximum probability; Represents a grid Generate observations The probability of; Denotes the initial state probability, and , Indicates the total number of navigation grids; Indicates the first moment of the grid The preceding state pointer; Indicates the first moment of the grid The preceding state pointer is null; These represent the ship trajectory observation value and the UAV trajectory observation value at the first moment, respectively; S2522, Calculation arrive Maximum probability and previous state pointer: ; ; In the formula, and They represent Time and Always in the grid The maximum probability; express Time Grid The set of adjacent navigation grids; express Time Grid Generate observations The probability of; express Time Grid The optimal preceding grid; Indicates the last moment; S2523, Determine the final moment Maximum probability grid And regard it as the endpoint of the optimal association path: ; In the formula, This represents the total probability of the optimal associated path; Indicates the last moment The maximum probability of all grids; S2524, From the endpoint of the optimal association path Triggered by the preceding state pointer Working backwards from each moment The optimal grid is obtained by arranging the reversed paths in forward order, resulting in the optimal grid sequence from the first time step to the last time step. This optimal grid sequence is then considered the optimal associated path. ; S2525, Calculate the matching confidence level : ; In the formula, This represents the sum of the maximum probabilities of all navigable grids at the final moment.

[0016] S3. Fuzzy Risk Factor Quantification and Modeling: Based on the optimal correlation path, the dynamic risks of ships, drones and collaborative interactions are quantified, and deep reinforcement learning is used to dynamically optimize the risk weights, outputting a comprehensive fuzzy risk value and risk level.

[0017] Step S3 specifically includes the following steps: S31. Based on the ship motion model and CFD performance correction, quantify the dynamic risks of the ship caused by environmental disturbances and control deviations, and obtain the total dynamic risk of the ship. : ; in, ; ; In the formula, These represent the stall risk and heading deviation risk after CFD performance data correction, respectively. and All represent weights; Indicates the difference between the ship's actual speed and its planned speed. The deviation, and , Indicates the tonnage of the ship; Indicates the actual heading versus the planned heading deviation, ; Indicates the maximum permissible heading deviation; S32. Based on the UAV aerodynamic model and ECMWF meteorological data, the flight stability risk of the UAV caused by wind and waves is quantified to obtain the total dynamic risk of the UAV. : ; in, ; ; In the formula, and All represent weights; and These represent the risk of lift loss and the risk of attitude deviation, respectively. This indicates the actual lift of the drone, and , This indicates the ideal lift of the drone. , Indicates air density, Indicates wing area. Indicates the lift coefficient. Indicates the actual attitude angle and the stable attitude angle deviation, ; Indicates the maximum permissible attitude angle; S33. Based on the spatiotemporal synchronization of multi-graph associations, calculate the risk of collaborative interaction. : ; in, ; ; ; In the formula, , and These represent collision risk, communication risk, and mission deviation risk, respectively. This indicates the real-time 3D distance between the ship and the drone, and ; Indicates a safe distance; Indicates the distance ambiguity coefficient; and These represent the real-time communication latency and the maximum allowed communication latency, respectively. This indicates the actual relative position deviation, and , Indicates the relative position of the plan; Indicates the maximum permissible deviation; S34. Employing deep reinforcement learning to dynamically optimize the overall dynamic risk of ships. Total dynamic risk of drones and collaborative interaction risks The weights are used to achieve adaptive aggregation of risks and output a comprehensive fuzzy risk. Risk level: ; ; In the formula, , and All of these represent weights that have been dynamically optimized through deep reinforcement learning.

[0018] In step S34, the state space is dynamically optimized by deep reinforcement learning. Action space , Initial weights; reward function Designed as follows: ; In the formula, Indicates predicted risk, and ; Indicates actual risk; Train a deep reinforcement learning model using historical risk data until it converges, and outputs the result. , and .

[0019] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A maritime ship-UAV collaborative risk assessment method integrating fuzzy risk and multi-map association, characterized by: Includes the following steps: S1. Multi-source data preprocessing and standardization: Simultaneously collect raw ship AIS data, UAV flight data and environmental data to form a raw dataset, and clean, smooth, spatiotemporally interpolate and standardize the raw dataset to output a standardized dataset with a unified benchmark. S2. Multi-graph construction and cross-graph association: Based on standardized data, construct an adaptive grid map of navigation space, ship trajectory map, UAV trajectory map and environmental risk map, and associate the adaptive grid map of navigation space, ship trajectory map, UAV trajectory map and environmental risk map through a hidden Markov model, and output the optimal association path and matching confidence. S3. Fuzzy Risk Factor Quantification and Modeling: Based on the optimal correlation path, the dynamic risks of ships, drones and collaborative interactions are quantified, and deep reinforcement learning is used to dynamically optimize the risk weights, outputting a comprehensive fuzzy risk value and risk level.

2. The maritime ship-UAV collaborative risk assessment method based on fuzzy risk and multi-graph association as described in claim 1, characterized in that: The standardized dataset described in step S1 , These represent standardized ship AIS data, drone flight data, and environmental data, respectively; among them, , These represent the ship's longitude, latitude, altitude, speed, heading, and timestamp, respectively. , These represent the longitude, latitude, altitude, speed, attitude angle, sensor noise variance, and timestamp of the UAV, respectively; environmental data. ECMWF meteorological data were analyzed using spatiotemporal bilinear interpolation. Discrete data is converted into continuous grid data and then fused with CFD performance data. Acquisition of ECMWF meteorological data , Representing wave height, wave direction, and wave period respectively, CFD performance data. , These represent the ship stall coefficient and the UAV lift loss coefficient, respectively. Spatiotemporal bilinear interpolation includes spatial bilinear interpolation and temporal linear interpolation: ； ； In the formula, Indicates the target location Interpolation results of environmental parameters at the location; express Any parameter in; Indicates surrounding Environmental parameter values ​​for the four selected ECMWF grid points; They represent the positions around the target location. ECMWF grid points were obtained; Indicates the target time Interpolation results of environmental parameters at the location; and They represent respectively; and Representing the target time respectively Interpolation results of environmental parameters at two different time points; Fusion CFD performance data The process is as follows: First, establish the ship type and ship stall coefficient respectively. And the drone model and the drone lift loss coefficient A mapping table between them is used to determine the lift loss coefficient of the UAV based on the current ship type and UAV model. and ship stall coefficient Finally, the lift loss coefficient of the drone was calculated. and ship stall coefficient Assign the value to the corresponding ECMWF mesh.

3. The maritime ship-UAV collaborative risk assessment method based on fuzzy risk and multi-graph association as described in claim 2, characterized in that: Step S2 specifically includes the following steps: S21. Construct an adaptive grid map of the navigation space: Use an adaptive quadtree to divide the collaborative operation area, in the ship-UAV interaction area. Perform grid division to generate an adaptive grid map of navigation space. ;in, This indicates the latitude and longitude coordinates of the lower left corner of the ship-drone collaborative operation area. This indicates the latitude and longitude coordinates of the upper right corner of the ship-drone collaborative operation area; Represents grid encoding; Indicates the grid type, and ; Indicates the coordinates of the grid center, and , and Let represent the grid side lengths of the adaptive quadtree in the longitude and latitude directions, respectively. , Indicates the depth of the partition; S22. Constructing a ship trajectory map: First, the ship's trajectory points Determine the matching grid by coordinate attribution: ； In the formula, The grid to which the ship's trajectory point belongs; Then connect the consecutive trajectory points in chronological order. If two adjacent ship trajectory points and The grid encoding satisfies Then a trajectory segment is generated. ; in, They represent the first The and the first The grid code to which each ship's trajectory point belongs; They represent the first The and the first Timestamps of each ship's trajectory point; This represents the average velocity of the trajectory segment, and , and They represent the first The and the first The actual speed of each ship's trajectory point; Finally, statistics for each grid. Number of trajectory segments : ； In the formula, Indicates an indicator function; and Representing trajectory segments The starting and ending grids; Represents trajectory segment A set; Obtain ship trajectory map ; S23. Construct a drone trajectory map; First, the drone's trajectory points Match to the corresponding grid: ； In the formula, The grid to which the drone's trajectory points belong; Then connect the consecutive trajectory points in chronological order. If two adjacent trajectory points... and The grid encoding satisfies Then a trajectory segment is generated. ; in, This represents the average velocity of the trajectory segment, and , and They represent the first The and the first The actual speed of each drone trajectory point; Indicates the average flight altitude of the trajectory segment; Finally, statistics for each grid. Number of trajectory segments : ； In the formula, Indicates an indicator function; Represents trajectory segment A set; Obtain ship trajectory map ; S24. Constructing an environmental risk map: Integrating standardized environmental data with an adaptive grid map of airspace. By correlating and quantifying the environmental threat level of each grid, an environmental risk map can be constructed. S241. Calculate the comprehensive wind and wave risk : ； in, ,and ; ； ； ,and ; ； In the formula, and Indicates the collaborative weight; and These respectively represent the wind and wave risks for ships and the wind and wave risks for drones; This indicates the rate of speed loss of a ship due to wind and waves; Indicates the ship's maximum speed; Indicates the angle between the ship's heading and the wave direction; Indicates the lift loss rate of the drone; This indicates the angle between the drone's attitude angle and the wave direction; S242. Calculate No-Fly and No-Sail Risks: Set no-fly and no-sail risks for no-fly and no-sail grids. Regarding the risks of no-fly and no-fly zones in the general aviation grid system. ; S243, Taking into account comprehensive wind and wave risks Risks of no-fly and no-fly zones The maximum value is used as the total grid risk value. : ； S25. Cross-graph association: Connect the adaptive grid map of navigation space, ship trajectory map, UAV trajectory map and environmental risk map through hidden Markov model, and output the optimal association path with the grid as the hidden state and the trajectory point as the observation value. S251. Define a Hidden Markov Model: Define the hidden state space. Adaptive mesh map for airspace Navigation grid in the observation space; Elements in the state transition matrix , Represents a grid Transfer to grid The probability, and , Indicates the risk discount factor. Represents a grid The number of adjacent navigation grids; elements in the observation matrix , Represents a grid Generate observations The probability of; Calculate the observation-grid matching probability : ； in, ； ； ； ； In the formula, Indicates weight; This indicates the probability of a match between ship observations and the grid. This indicates the probability of a match between UAV observations and the grid. This represents the distance from the ship's trajectory point to the center of the grid. Indicates the ship's positioning error; This represents the distance from the drone's trajectory point to the center of the grid. This indicates the drone's positioning error; S252. Use the Viterbi algorithm to solve for the optimal association path and confidence of the Hidden Markov Model, and obtain the multi-graph association results. , Indicates the optimal association path. Indicates the match confidence level; S253, Determine whether the condition is satisfied. If yes, proceed to step S3; otherwise, return to step S21.

4. The maritime ship-UAV collaborative risk assessment method based on fuzzy risk and multi-graph association as described in claim 3, characterized in that: In step S21, the mesh type determination strategy is as follows: If the grid node is entirely within a no-fly or no-navigation zone: ; If the grid nodes are entirely within the navigable area: ; If grid nodes partially overlap, then mark them as grid nodes to be subdivided. For a node to be subdivided, further subdivision to child nodes is allowed if one of the following conditions is met: Ship-UAV trajectory density within nodes , Indicates the number of ship-drone trajectory segments; Indicates the grid area; This indicates setting a trajectory density threshold; Distance between the node and the boundary of the no-fly or no-navigation zone , This indicates setting a boundary distance threshold; Furthermore, the subdivided child node encoding rules are as follows: assuming the parent node is encoded as... The codes for the four child nodes are as follows: .

5. The maritime ship-UAV collaborative risk assessment method based on fusion of fuzzy risk and multi-graph association as described in claim 4, characterized in that: Step S252 specifically includes the following steps: S2521. Initialization: Initialize all navigation grids at the first moment to generate the first observation. The maximum probability, and the previous state pointer: ； ； In the formula, Indicates that the first moment is in the grid The maximum probability; Represents a grid Generate observations The probability of; Denotes the initial state probability, and , Indicates the total number of navigation grids; Indicates the first moment of the grid The preceding state pointer; Indicates the first moment of the grid The preceding state pointer is null; These represent the ship trajectory observation value and the UAV trajectory observation value at the first moment, respectively; S2522, Calculation arrive Maximum probability and previous state pointer: ； ； In the formula, and They represent Time and Always in the grid The maximum probability; express Time Grid The set of adjacent navigation grids; express Time Grid Generate observations The probability of; express Time Grid The optimal preceding grid; Indicates the last moment; S2523, Determine the final moment Maximum probability grid And regard it as the endpoint of the optimal association path: ； In the formula, This represents the total probability of the optimal associated path; Indicates the last moment The maximum probability of all grids; S2524, From the endpoint of the optimal association path Triggered by the preceding state pointer Working backwards from each moment The optimal grid is obtained by arranging the reversed paths in forward order, resulting in the optimal grid sequence from the first time step to the last time step. This optimal grid sequence is then considered the optimal associated path. ; S2525, Calculate the matching confidence level : ； In the formula, This represents the sum of the maximum probabilities of all navigable grids at the final moment.

6. The maritime ship-UAV collaborative risk assessment method based on fusion of fuzzy risk and multi-graph association as described in claim 5, characterized in that: Step S3 specifically includes the following steps: S31. Based on the ship motion model and CFD performance correction, quantify the dynamic risks of the ship caused by environmental disturbances and control deviations, and obtain the total dynamic risk of the ship. : ； in, ； ； In the formula, These represent the stall risk and heading deviation risk after CFD performance data correction, respectively. and All represent weights; Indicates the difference between the ship's actual speed and its planned speed. The deviation, and , Indicates the tonnage of the ship; Indicates the actual heading versus the planned heading deviation, ; Indicates the maximum permissible heading deviation; S32. Based on the UAV aerodynamic model and ECMWF meteorological data, the flight stability risk of the UAV caused by wind and waves is quantified to obtain the total dynamic risk of the UAV. : ； in, ； ； In the formula, and All represent weights; and These represent the risk of lift loss and the risk of attitude deviation, respectively. This indicates the actual lift of the drone, and , This indicates the ideal lift of the drone. , Indicates air density, Indicates wing area. Indicates the lift coefficient. Indicates the actual attitude angle and the stable attitude angle deviation, ; Indicates the maximum permissible attitude angle; S33. Based on the spatiotemporal synchronization of multi-graph associations, calculate the risk of collaborative interaction. : ； in, ； ； ； In the formula, , and These represent collision risk, communication risk, and mission deviation risk, respectively. This indicates the real-time 3D distance between the ship and the drone, and ; Indicates a safe distance; Indicates the distance ambiguity coefficient; and These represent the real-time communication latency and the maximum allowed communication latency, respectively. This indicates the actual relative position deviation, and , Indicates the relative position of the plan; Indicates the maximum permissible deviation; S34. Employing deep reinforcement learning to dynamically optimize the overall dynamic risk of ships. Total dynamic risk of drones and collaborative interaction risks The weights are used to achieve adaptive aggregation of risks and output a comprehensive fuzzy risk. Risk level: ； ； In the formula, , and All of these represent weights that have been dynamically optimized through deep reinforcement learning.

7. The maritime ship-UAV collaborative risk assessment method based on fusion of fuzzy risk and multi-graph association as described in claim 6, characterized in that: In step S34, the state space is dynamically optimized by deep reinforcement learning. Action space , Indicates the initial weights; reward function Designed as follows: ； In the formula, Indicates predicted risk, and ; Indicates actual risk; Train a deep reinforcement learning model using historical risk data until it converges, and outputs... , and .

Images