A multi-ship cooperative collision avoidance decision method and system considering risk coupling

Abstract

This invention provides a multi-ship cooperative collision avoidance decision-making method and system considering risk coupling, belonging to the field of intelligent ship autonomous navigation technology. The method includes: acquiring the position, speed, heading, length, and safety domain parameters of each ship in a multi-ship encounter scenario; calculating relative motion parameters such as DCPA and TCPA; integrating objective collision risk, circular ship domain constraints, and prospect theory, and constructing a risk distance matrix using negative logarithmic mapping; identifying local risk clusters through OPTICS, and determining collision avoidance priorities based on cluster-level risk and global ship risk; using collision risk, range loss, turning radius, avoidance responsibility, and safety distance as evaluation indicators, and employing AHP and an improved differential evolution algorithm incorporating chaotic initialization, parallel subpopulation evolution, inferior individual elimination, elite retention, and parameter adaptation to solve the cooperative collision avoidance scheme in stages. This invention can characterize the risk coupling relationship among multiple ships and generate cooperative collision avoidance decisions that satisfy safety distances and international maritime collision avoidance rules.

Description

Technical Field

[0001] This invention relates to the fields of intelligent ship autonomous navigation, ship collision risk assessment and multi-ship cooperative control technology, and more specifically, to a multi-ship cooperative collision avoidance decision-making method and system that considers risk coupling. Background Technology

[0002] In ports, coastal waters, and important waterways, vessel traffic density is constantly increasing, and the relationships between vessels in multi-vessel encounter scenarios exhibit dynamic changes, mutual influences, and risk coupling. A vessel's local collision avoidance maneuver can alter its relative motion with multiple surrounding vessels, further changing the risk structure of the entire encounter scenario. Therefore, addressing collision hazards independently on a pair-by-pair basis makes it difficult to directly obtain a collaborative collision avoidance solution that balances overall safety and feasibility.

[0003] Existing ship collision avoidance methods typically use DCPA, TCPA, real-time distance, relative bearing, or relative speed to establish an objective collision risk level and generate turning or path planning results based on international maritime collision avoidance rules. These methods can describe the geometric proximity of ship pairs, but they are insufficient in describing the nonlinear risk perception when the safety domain is intruded and the amplification effect of low-probability, high-consequence risks. Therefore, they may miss local high-risk relationships in dense multi-ship scenarios.

[0004] On the other hand, clustering or conflict clustering methods for multi-ship scenarios can identify local high-risk ship groups, but if the clustering distance is only composed of spatial distance or objective risk level, it is easily affected by the sparse distribution of risk, uneven density and threshold selection. The identified risk clusters need to be further transformed into cluster-level and ship-level collision avoidance priorities and connected with the action optimization process that meets the rules, safe distance and voyage economy.

[0005] Therefore, a method is needed that can integrate objective collision risk level, ship domain constraints and subjective risk perception, identify multi-ship risk coupling structure, and update the encounter state and solve cooperative collision avoidance actions for each ship according to risk priority. Summary of the Invention

[0006] To address the technical problem that existing technologies struggle to simultaneously describe the risk coupling relationships among multiple vessels, determine local risk clusters and key vessel priorities, and generate rule-compatible cooperative collision avoidance maneuvers, this invention provides a multi-vessel cooperative collision avoidance decision-making method and system that considers risk coupling. This invention primarily utilizes prospect theory to perform nonlinear correction on objective collision risks and employs negative logarithmic mapping to construct a risk distance matrix to identify local risk clusters. Furthermore, it combines cluster-level and vessel-level dual priorities with an improved differential evolution algorithm to solve for cooperative collision avoidance maneuvers in stages. This approach effectively characterizes the risk coupling relationships among multiple vessels and generates cooperative collision avoidance decisions that satisfy safety distances and international maritime collision avoidance rules.

[0007] The technical means employed in this invention are as follows: A multi-ship cooperative collision avoidance decision-making method considering risk coupling includes: S1. Obtain the position, speed, heading, length, and safety parameters of each ship in a multi-ship encounter scenario. S2. Based on the obtained positions, speeds and headings of each vessel, calculate the relative motion parameters between each pair of vessels and the safe distance threshold between the pairs of vessels. S3. Calculate the objective collision risk based on relative motion parameters, determine the ship domain constraints by combining the ship pair safety distance threshold, and use prospect theory to perform nonlinear correction on the objective collision risk to construct a subjective-objective integrated risk value. S4. Symmetricize, normalize, and perform negative logarithmic mapping on the constructed subjective and objective integrated risk values ​​to construct a risk distance matrix; S5. Input the constructed risk distance matrix into the OPTICS clustering algorithm to generate an ordered ship sequence and reachability sequence through core distance and reachability distance. Extract continuous valley areas as local risk clusters based on the reachability distance threshold. S6. Based on the average risk at the cluster level, the maximum ship-to-ship risk within the cluster, and the global risk of each ship, determine the processing order of risk clusters and the processing order of key ships within the cluster to obtain the collaborative collision avoidance priority. S7. Construct a judgment matrix using collision risk, range loss, turning radius, avoidance responsibility, and safe distance as evaluation indicators, and use the analytic hierarchy process (AHP) to determine the weight of each evaluation indicator. S8. Based on the priority of collaborative collision avoidance, a comprehensive fitness function is constructed by weighting each evaluation index according to the determined weights. An improved differential evolution algorithm is used to solve the turning amplitude, collision avoidance start time and rerouting time of each ship to be decided in turn. After each ship's action is determined, the multi-ship encounter state is updated, and the multi-ship collaborative collision avoidance decision scheme is output.

[0008] Further, step S1 includes: S11. Receive multi-source data provided by automatic identification systems, radar, satellite positioning equipment and electronic chart systems; S12. Clean the multi-source data to form the time-lapse data for each ship. The state vector is as follows:

[0009] in, For position coordinates, , For speed, , For the course, As captain, For the radius of the safety zone.

[0010] Further, step S2 includes: S21. For any ship With ships The relative position vectors are constructed based on the obtained positions of each ship, as shown in the following formula:

[0011] in, Indicates a ship Location, Indicates a ship Location; S22. For any ship With ships The relative velocity vector is constructed based on the obtained speed and heading of each ship, as shown in the following formula:

[0012] in, Indicates a ship speed, Indicates a ship The speed of the ship; S23, Calculate real-time distance Relative velocity Relative orientation and speed ratio The formula is as follows:

[0013]

[0014]

[0015]

[0016] in, Used to map angles to a preset angle range. To prevent positive numbers with a denominator of zero; S24. Calculate the minimum meeting time and nearest meeting distance based on the constructed relative position vector and relative velocity vector, using the following formula:

[0017]

[0018] S25. Calculate the safety zone radius of each vessel based on the obtained ship length and safety zone parameters. And calculate the ship With ships The safe distance threshold between them is calculated using the following formula:

[0019] in, For safety scale coefficient, Here is the conversion factor between meters and nautical miles; S26. Identify the encounter, crossing, or overtaking relationship based on the calculated relative bearing and heading difference, and determine the yielding responsibility and permissible turning direction in accordance with the International Regulations for Preventing Collisions at Sea.

[0020] Further, step S3 includes: S31. Based on the calculated nearest encounter distance (DCPA), minimum encounter time (TCPA), real-time distance, relative orientation, and velocity ratio, calculate the objective collision risk using the following formula:

[0021] in, , , , and The weights are non-negative, and the sum of the weights is 1. , , , and For the corresponding normalized risk factor; S32. Determine the vessel domain constraints based on the safety domain radius and the obtained location, and establish a circular vessel domain around each vessel, as shown in the following formula:

[0022] S33. Employing prospect theory to perform nonlinear correction on objective collision risk, constructing a combined subjective and objective risk value, specifically including: The objective collision risk level is mapped to a probabilistic surrogate quantity to avoid numerical anomalies in the probability weighting function when the probability is 0 or 1:

[0023] The nonlinear perception of low-probability risks is described using a probability weighting function from prospect theory:

[0024] in, and These are probability weight parameters. This can enhance the perceived weight of low-probability risks; The distance loss and time loss are calculated based on the ship-to-ship safe distance threshold, using the following formula:

[0025]

[0026] in, and These are reference values ​​for safe distance and safe time, respectively. With safe distance threshold Maintain consistency; The severity of the risk is obtained by weighting the distance loss and time loss, as shown in the following formula:

[0027] in, and All are weighting coefficients, and ; Considering loss aversion and diminishing marginal sensitivity, a subjective value and a risk value combining subjective and objective factors are constructed, as follows:

[0028]

[0029] in, For the ship Subjective value, The loss aversion coefficient, Assess the severity of the risk. The sensitivity coefficient, Combine subjective and objective risk values. This is the probabilistic proxy quantity after mapping objective collision risk. The scale parameter of the Prelec function. This is the curvature parameter of the Prelec function.

[0030] Further, step S4 includes: S41. Symmetrically transform the constructed subjective-objective fusion risk value to obtain the symmetrical subjective-objective fusion risk, as shown in the following formula:

[0031] S42. Normalize using the maximum value of the non-diagonal elements within the scene to form... :

[0032] S43. Construct the risk distance using a negative logarithmic mapping, as shown in the following formula:

[0033] in, To prevent positive numbers from having zero values ​​during normalization and logarithmic operations, the diagonal elements are fixed to zero, thus obtaining the risk distance matrix.

[0034] Further, step S5 includes: S51, the negative logarithmic risk distance matrix As a pre-computed distance input for the OPTICS clustering algorithm; S52, For ships The core distance is the risk distance to the MinPts-th nearest neighbor, and the formula is as follows:

[0035] S53, Regarding current vessels and its neighboring vessels The reachable distance is calculated using the following formula:

[0036] S54 and OPTICS continuously update the reachability distances of each vessel, resulting in an ordered vessel sequence. and the corresponding reachability sequence ; S55. In the reachability distance curve, continuous valleys correspond to locally high-risk clusters, and peaks correspond to the separation between risk clusters; let the clustering distance threshold be... Extracting continuous satisfaction Sequence segments with at least MinPts of ships are considered local risk clusters. ; S56. The effectiveness of risk cluster identification is evaluated using coverage, Davies-Bouldin index (DBI), coverage penalty (DBI), and high-risk recall, as shown in the following formula:

[0037]

[0038]

[0039] Among them, the higher the coverage and high-risk recall, and the lower the DBI and coverage penalty DBI, the more reliable the local risk cluster identification results are.

[0040] Further, step S6 includes: S61. For localized risk clusters The cluster-level average risk and the maximum ship pair risk within the cluster are calculated using the following formulas:

[0041]

[0042] S62, Calculation of Ships The global risk in the entire multi-ship risk network is calculated as follows:

[0043] S63, First, calculate the average risk at the cluster level. Risk clusters are sorted in descending order. When the average risks of two risk clusters are close, the cluster with the highest ship-to-ship risk is selected. Prioritize larger risk clusters and determine the processing order for risk clusters; S64. For vessels within the same risk cluster, determine the key vessel sequence according to the global risk descending order, and combine this with encounter responsibility to exclude straight-running vessels that do not require active maneuvering, to obtain the cooperative collision avoidance priority.

[0044] Further, step S7 includes: S71. Construct a five-order judgment matrix using collision risk, range loss, turning angle, avoidance responsibility, and safe distance as evaluation indicators. ; S72, Solving the judgment matrix The largest eigenvalue and its corresponding eigenvector are obtained, and the eigenvector is normalized to obtain the weight vector. ; S73. Calculate the consistency ratio and perform a consistency test. The formula is as follows:

[0045] in, As the average random consistency index, when If the value is less than 0.1, the weight vector is accepted; otherwise, the judgment matrix is ​​adjusted and the calculation is repeated.

[0046] Further, step S8 includes: S81. Calculate the collision risk assessment item based on the overall ship risk after adopting candidate maneuvers. The formula is as follows:

[0047] S82. Calculate the distance loss item to reflect the cost of detour. The formula is as follows:

[0048] S83. Calculate the steering amplitude term used to reflect the steering amplitude. The formula is as follows:

[0049] S84. Calculate the avoidance liability assessment item that includes the International Maritime Collision Prevention Regulations and penalties for violations of the give-way duty. The formula is as follows:

[0050] Among them, when the candidate maneuver simultaneously meets the requirements of the International Maritime Collision Prevention Regulations and the obligation to avoid collision. If the value is 0, otherwise a positive penalty value is applied. ; S85. Calculate the safety distance evaluation term for candidate schemes whose closest distance is less than the ship-to-ship safety distance threshold. The formula is as follows:

[0051] in, For safe distance threshold, ; S86. The multi-stage cooperative collision avoidance problem is represented as a multi-objective constrained optimization problem. The five evaluation indicators are unified as collision risk, range loss, turning angle, avoidance responsibility, and safe distance. The comprehensive fitness function is then constructed as follows:

[0052] in, , , , , All are weighting coefficients; S87, Using a chaotic mapping function The initial individuals are generated within the boundaries of each decision variable, as shown in the following formula:

[0053]

[0054]

[0055] in, For chaotic parameters, the generated initial individuals cover the full range of decision variables; S88. Divide the overall population into Each subpopulation performs differential mutation in parallel within the same generation; individuals in each generation are sorted according to their fitness function values, and superior individuals are written into the elite archive and allowed to be used as mutation basis vectors or to replace inferior individuals; individuals that are consistently in the inferior range are eliminated or chaotically reinitialized to maintain population diversity. S89. Perform crossover and greedy selection, retaining experimental individuals with fitness no worse than their parents. The corresponding population partitioning, mutation, crossover, and selection relationships are as follows:

[0056]

[0057] The above formula employs a general bounded update framework for success parameter statistics. and Each by the first The scaling factor and crossover factor are calculated to generate better experimental individuals. and For update rate, Limit the results to a given range; S810. Execute decisions ship by ship according to the determined cooperative collision avoidance priority: For the current critical ship, use the constructed comprehensive fitness function as the objective to solve for the turning amplitude, collision avoidance start time, and return-to-course time, while maintaining the current navigation state of the remaining undecided ships. The decision vector for each critical ship is defined as:

[0058] in, For the steering range, To avoid collision at the start time, The time for resumption of service; S811. After writing the obtained actions into the scene state, recalculate the relative motion parameters of all ship pairs and the safe distance threshold between ship pairs and update the encounter state. S812. Solve for the next critical vessel. Repeat steps S810 to S81 until the collaborative collision avoidance decision for all critical vessels is completed. Output the turning amplitude, collision avoidance start time, rerouting time and reference trajectory of each vessel.

[0059] This invention also provides a multi-ship cooperative collision avoidance decision-making system considering risk coupling, implemented based on the above-mentioned multi-ship cooperative collision avoidance decision-making method considering risk coupling, comprising: a state acquisition module, a relative motion parameter calculation module, a risk fusion module, a risk distance matrix construction module, a risk cluster identification module, a collision avoidance priority determination module, a weight determination module, and a cooperative optimization and decision output module, wherein: The status acquisition module is used to acquire the position, speed, heading, length and safety parameters of each ship in a multi-ship encounter scenario. The relative motion parameter calculation module is used to calculate the relative motion parameters between each pair of ships and the safe distance threshold between the pairs of ships based on the obtained positions, speeds and headings of each ship. The risk fusion module is used to calculate the objective collision risk based on relative motion parameters, determine the ship domain constraints by combining the ship pair safety distance threshold, and use prospect theory to perform nonlinear correction on the objective collision risk to construct a subjective-objective fusion risk value. The risk distance matrix construction module is used to symmetrize, normalize, and perform negative logarithmic mapping on the constructed subjective and objective integrated risk values ​​to construct a risk distance matrix. The risk cluster identification module is used to input the constructed risk distance matrix into the OPTICS clustering algorithm, generate an ordered ship sequence and reachability sequence through core distance and reachability distance, and extract continuous valley areas as local risk clusters based on the reachability distance threshold; The collision avoidance priority determination module is used to determine the processing order of risk clusters and the processing order of key ships within clusters based on the extracted cluster-level average risk of each local risk cluster, the maximum ship-to-ship risk within the cluster, and the global risk of each ship, so as to obtain the collaborative collision avoidance priority. The weight determination module is used to construct a judgment matrix with collision risk, range loss, turning radius, avoidance responsibility and safe distance as evaluation indicators, and to determine the weight of each evaluation indicator using the analytic hierarchy process. The collaborative optimization and decision output module is used to construct a comprehensive fitness function by weighting each evaluation index according to the collaborative collision avoidance priority and a determined weight. It uses an improved differential evolution algorithm to solve the turning amplitude, collision avoidance start time and rerouting time of each ship to be decided in turn. After each ship's action is determined, it updates the multi-ship encounter state and outputs the multi-ship collaborative collision avoidance decision scheme.

[0060] Compared with the prior art, the present invention has the following advantages: 1. The present invention provides a multi-ship collaborative collision avoidance decision-making method that considers risk coupling. By performing nonlinear correction on objective collision risk with prospect theory and constructing a risk distance matrix using negative logarithmic mapping, it realizes the accurate description of nonlinear risk perception when the safety domain is invaded and the amplification effect of low probability high consequences, and enhances the distinguishability of high-risk ship pairs in the clustering distance space.

[0061] 2. The present invention provides a multi-ship collaborative collision avoidance decision-making method that considers risk coupling. By inputting the risk distance matrix into the OPTICS clustering algorithm to identify local risk clusters, and combining the cluster-level average risk, the maximum ship pair risk within the cluster and the global ship risk, the processing order of risk clusters and the processing order of key ships within the cluster are determined. This achieves accurate extraction and hierarchical priority ranking of local high-risk ship groups in multi-ship encounter scenarios with uneven density and sparse risk distribution.

[0062] 3. The present invention provides a multi-ship collaborative collision avoidance decision-making method that considers risk coupling. By solving the turning amplitude, collision avoidance start time and rerouting time for each ship according to the collaborative collision avoidance priority, and updating the multi-ship encounter state after each ship's action is determined, a phased collaborative collision avoidance decision-making is realized, which effectively reduces the decision-making conflict caused by simultaneously optimizing all ships.

[0063] 4. The present invention provides a multi-ship cooperative collision avoidance decision-making method that considers risk coupling. By adopting an improved differential evolution algorithm that includes chaotic initialization, parallel subpopulation evolution, elimination of inferior individuals, preservation of elites and parameter adaptation, it achieves the goal of avoiding the loss of high-quality solutions while maintaining population diversity, thereby improving the solution accuracy and convergence stability of the cooperative collision avoidance scheme.

[0064] 5. The present invention provides a multi-ship collaborative collision avoidance decision-making method that considers risk coupling. By constructing a comprehensive fitness function with collision risk, range loss, turning radius, avoidance responsibility and safe distance as evaluation indicators, and using the analytic hierarchy process to determine the weights, it achieves a comprehensive balance of safety, rule compatibility and range economy.

[0065] In summary, by applying the technical solution of this invention, existing technologies only handle collision hazards independently for each ship pair, which is insufficient in describing the nonlinear risk perception when the safety domain is intruded upon and the amplification effect of low-probability, high-consequence risks. They also struggle to identify unevenly distributed local high-risk relationships, and optimizing all ships can easily lead to decision-making conflicts. This invention uses foreground theory to perform nonlinear correction on objective collision risks, employs negative logarithmic mapping to construct a risk distance matrix to identify local risk clusters, and combines cluster-level and ship-level dual priority with an improved differential evolution algorithm to solve cooperative collision avoidance actions in stages. Therefore, the technical solution of this invention solves the problem in existing technologies of being unable to simultaneously describe the risk coupling relationships of multiple ships, determine local risk clusters and the priority of key ships, and generate rule-compatible cooperative collision avoidance actions.

[0066] Based on the above reasons, this invention can be widely applied in fields such as intelligent ship autonomous navigation, ship collision risk assessment, multi-ship collaborative control, port and coastal water traffic management, and ship-shore collaborative computing platforms. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is a flowchart illustrating the overall process of the multi-ship cooperative collision avoidance decision-making method of the present invention. Figure 2 This is a schematic diagram illustrating the construction of the multi-ship risk coupling relationship and risk distance matrix in this invention; Figure 3 This is a flowchart of the collaborative collision avoidance solution based on the analytic hierarchy process and the improved differential evolution algorithm of this invention. Figure 4This is a schematic diagram of the multi-ship cooperative collision avoidance decision system of the present invention; Figure 5 This is a schematic diagram illustrating the dynamic evolution of the risk structure in a twenty-ship encounter scenario according to an embodiment of the present invention. Figure 6 This is a comparison diagram of the risk matrix and reachable distance threshold segmentation in a 20-ship encounter scenario according to an embodiment of the present invention; Figure 7 This is a comparison chart of the risk cluster identification results before and after introducing prospect theory in step 90 of this embodiment of the invention; Figure 8 This is a cluster-level risk and global ship risk distribution diagram in step 90 of this embodiment of the invention; Figure 9 This is a diagram showing the risk cluster identification results of a ten-ship encounter scenario in an embodiment of the present invention. Figure 10 This is a collaborative collision avoidance trajectory diagram for a ten-ship encounter scenario in an embodiment of the present invention. Detailed Implementation

[0069] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0070] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product or device.

[0071] The following terms and units are used uniformly in this invention: the total number of ships in the scenario is Ships and ships Represents any different vessel; position and distance between vessels are in nautical miles, speed is in knots (nautical miles per hour), and vessel length is originally in meters, converted to nautical miles by dividing by 1852 when entering the distance model; TCPA and safe time thresholds. Use hours; minutes should be converted to hours first. Simulation time is calculated from the simulation steps and corresponding step sizes. Heading and relative bearing are in degrees; convert them to radians before substituting them into trigonometric functions. The coordinate system is based on due east. The positive axis and the due north direction are Positive axis, heading angle Starting with true north as 0° and increasing clockwise. Indicates objective collision risk. This represents the combined subjective and objective risk value after correction based on ship industry constraints and prospect theory. This indicates the risk of symmetric fusion. This represents normalized symmetric risk. Indicates the negative logarithmic risk distance; and All are numerically stable positive numbers, among which It is specifically designed for normalization and logarithmic operations.

[0072] like Figure 1 As shown, this invention provides a multi-ship cooperative collision avoidance decision-making method considering risk coupling, including: S1. Obtain the position, speed, heading, length, and safety parameters of each ship in a multi-ship encounter scenario. S2. Based on the obtained positions, speeds and headings of each vessel, calculate the relative motion parameters between each pair of vessels and the safe distance threshold between the pairs of vessels. S3. Calculate the objective collision risk based on relative motion parameters, determine the ship domain constraints by combining the ship pair safety distance threshold, and use prospect theory to perform nonlinear correction on the objective collision risk to construct a subjective-objective integrated risk value. S4. Symmetricize, normalize, and perform negative logarithmic mapping on the constructed subjective and objective integrated risk values ​​to construct a risk distance matrix; S5. Input the constructed risk distance matrix into the OPTICS clustering algorithm to generate an ordered ship sequence and reachability sequence through core distance and reachability distance. Extract continuous valley areas as local risk clusters based on the reachability distance threshold. S6. Based on the average risk at the cluster level, the maximum ship-to-ship risk within the cluster, and the global risk of each ship, determine the processing order of risk clusters and the processing order of key ships within the cluster to obtain the collaborative collision avoidance priority. S7. Construct a judgment matrix using collision risk, range loss, turning radius, avoidance responsibility, and safe distance as evaluation indicators, and use the analytic hierarchy process (AHP) to determine the weight of each evaluation indicator. S8. Based on the priority of collaborative collision avoidance, a comprehensive fitness function is constructed by weighting each evaluation index according to the determined weights. An improved differential evolution algorithm is used to solve the turning amplitude, collision avoidance start time and rerouting time of each ship to be decided in turn. After each ship's action is determined, the multi-ship encounter state is updated, and the multi-ship collaborative collision avoidance decision scheme is output.

[0073] In a specific implementation, as a preferred embodiment of the present invention, step S1 includes: S11. Receive multi-source data provided by automatic identification systems, radar, satellite positioning equipment and electronic chart systems; S12. Clean the multi-source data to form the time-lapse data for each ship. The state vector is as follows:

[0074] in, For position coordinates, , For speed, , For the course, As captain, For the radius of the safety zone.

[0075] In a specific implementation, as a preferred embodiment of the present invention, step S2 includes: S21. For any ship With ships The relative position vectors are constructed based on the obtained positions of each ship, as shown in the following formula:

[0076] in, Indicates a ship Location, Indicates a ship Location; S22. For any ship With ships The relative velocity vector is constructed based on the obtained speed and heading of each ship, as shown in the following formula:

[0077] in, Indicates a ship speed, Indicates a ship The speed of the ship; S23, Calculate real-time distance Relative velocity Relative orientation and speed ratio The formula is as follows:

[0078]

[0079]

[0080]

[0081] in, Used to map angles to a preset angle range. To prevent positive numbers with a denominator of zero; S24. Calculate the minimum meeting time and nearest meeting distance based on the constructed relative position vector and relative velocity vector, using the following formula:

[0082]

[0083] S25. Calculate the safety zone radius of each vessel based on the obtained ship length and safety zone parameters. And calculate the ship With ships The safe distance threshold between them is calculated using the following formula:

[0084] in, For safety scale coefficient, The conversion factor between meters and nautical miles; in this embodiment, the invention uniformly determines the safety zone radius based on the dimensions of each ship, and uses... As a reference value for safe distance between ships, another set of safe radii is no longer defined simultaneously using the maximum ship length in the scenario. The 0.9, 1.0, and 1.1 nautical miles directly listed in the ten-ship embodiments are values ​​obtained from the embodiments based on ship dimensions, safety dimension coefficients, or preset safety domain parameters.

[0085] S26. Identify the encounter, crossing, or overtaking relationship based on the calculated relative bearing and heading difference, and determine the yielding responsibility and permissible turning direction in accordance with the International Regulations for Preventing Collisions at Sea.

[0086] Table 1 Typical Encounter Relationships and Manipulation Requirements

[0087] In a specific implementation, as a preferred embodiment of the present invention, step S3 includes: S31. Based on the calculated nearest encounter distance (DCPA), minimum encounter time (TCPA), real-time distance, relative orientation, and velocity ratio, calculate the objective collision risk using the following formula:

[0088] in, , , , and The weights are non-negative, and the sum of the weights is 1. , , , and For the corresponding normalized risk factor; in this embodiment, such as Figure 2 As shown, the objective collision hazard level considers DCPA, TCPA, real-time distance, relative bearing, and velocity ratio. A monotonically normalized function is used to map each factor to... For intervals, the higher the degree of danger, the larger the normalized value. The smaller the DCPA or real-time distance, the larger the distance risk factor; when TCPA is positive and close to the current moment, the time risk factor is larger; the relative azimuth and speed ratio risk factors are determined according to the encounter relationship and relative motion trend.

[0089] S32. Determine the ship domain constraints based on the safety domain radius and the obtained location. To explicitly represent the differences in ship dimensions and the safety boundary, a circular ship domain is set around each ship, as shown in the following formula:

[0090] S33. Employing prospect theory to perform nonlinear correction on objective collision risk, constructing a combined subjective and objective risk value, specifically including: The objective collision risk level is mapped to a probabilistic surrogate quantity to avoid numerical anomalies in the probability weighting function when the probability is 0 or 1:

[0091] The nonlinear perception of low-probability risks is described using a probability weighting function from prospect theory:

[0092] in, and These are probability weight parameters. This can enhance the perceived weight of low-probability risks; The distance loss and time loss are calculated based on the ship-to-ship safe distance threshold, using the following formula:

[0093]

[0094] in, and These are reference values ​​for safe distance and safe time, respectively. With safe distance threshold To maintain consistency; in this embodiment, the above formula limits both distance loss and time loss to within... Within the range. When , Alternatively, when the inner product of relative position and relative velocity is not less than 0, indicating that the vessel is in a state of distance, the time loss is 0, in order to avoid identifying a pair of vessels that has already passed the nearest meeting point as a continuous approach.

[0095] The severity of the risk is obtained by weighting the distance loss and time loss, as shown in the following formula:

[0096] in, and All are weighting coefficients, and ; Considering loss aversion and diminishing marginal sensitivity, a subjective value and a risk value combining subjective and objective factors are constructed, as follows:

[0097]

[0098] in, For the ship Subjective value, The loss aversion coefficient, Assess the severity of the risk. The sensitivity coefficient, Combine subjective and objective risk values. This is the probabilistic proxy quantity after mapping objective collision risk. The scale parameter of the Prelec function. This is the curvature parameter of the Prelec function.

[0099] In a specific implementation, as a preferred embodiment of the present invention, step S4 includes: S41. Symmetrically transform the constructed subjective-objective fusion risk value to obtain the symmetrical subjective-objective fusion risk, as shown in the following formula:

[0100] In this embodiment, symmetry ensures that OPTICS receives symmetrical ship pair distances.

[0101] S42. Normalize using the maximum value of the non-diagonal elements within the scene to form... :

[0102] S43. Construct the risk distance using a negative logarithmic mapping, as shown in the following formula:

[0103] in, To prevent positive numbers with zero values ​​from being generated during normalization and logarithmic operations, the diagonal elements are fixed to zero, thus obtaining the risk distance matrix. In this embodiment, the negative logarithmic risk distance is used to convert the fused risk matrix into a pre-calculated distance input for OPTICS: the higher the ship-pair fusion risk, the closer the normalized risk is to 1, and the smaller the resulting risk distance; the lower the ship-pair fusion risk, the larger the resulting risk distance.

[0104] In a specific implementation, as a preferred embodiment of the present invention, step S5 includes: S51, the negative logarithmic risk distance matrix As a pre-computed distance input for the OPTICS clustering algorithm; S52, For ships The core distance is the risk distance to the MinPts-th nearest neighbor, and the formula is as follows:

[0105] S53, Regarding current vessels and its neighboring vessels The reachable distance is calculated using the following formula:

[0106] S54 and OPTICS continuously update the reachability distances of each vessel, resulting in an ordered vessel sequence. and the corresponding reachability sequence ; S55. In the reachability distance curve, continuous valleys correspond to locally high-risk clusters, and peaks correspond to the separation between risk clusters; let the clustering distance threshold be... Extracting continuous satisfaction Sequence segments with at least MinPts of ships are considered local risk clusters. ; S56. The effectiveness of risk cluster identification is evaluated using coverage, Davies-Bouldin index (DBI), coverage penalty (DBI), and high-risk recall, as shown in the following formula:

[0107]

[0108]

[0109] Among them, the higher the coverage and high-risk recall, and the lower the DBI and coverage penalty DBI, the more reliable the local risk cluster identification results are.

[0110] In a specific implementation, as a preferred embodiment of the present invention, step S6 includes: S61. For localized risk clusters The cluster-level average risk and the maximum ship pair risk within the cluster are calculated using the following formulas:

[0111]

[0112] S62, Calculation of Ships The global risk in the entire multi-ship risk network is calculated as follows:

[0113] S63, First, calculate the average risk at the cluster level. Risk clusters are sorted in descending order. When the average risks of two risk clusters are close, the cluster with the highest ship-to-ship risk is selected. Prioritize larger risk clusters and determine the processing order for risk clusters; S64. For vessels within the same risk cluster, determine the key vessel sequence according to the global risk descending order, and combine this with encounter responsibility to exclude straight-running vessels that do not require active maneuvering, to obtain the cooperative collision avoidance priority.

[0114] In specific implementation, as a preferred embodiment of the present invention, such as Figure 3 As shown, step S7 includes: S71. Construct a five-order judgment matrix using collision risk, range loss, turning angle, avoidance responsibility, and safe distance as evaluation indicators. ; S72, Solving the judgment matrix The largest eigenvalue and its corresponding eigenvector are obtained, and the eigenvector is normalized to obtain the weight vector. ; S73. Calculate the consistency ratio and perform a consistency test. The formula is as follows:

[0115] in, As the average random consistency index, when If the value is less than 0.1, the weight vector is accepted; otherwise, the judgment matrix is ​​adjusted and recalculated. In this embodiment, the obtained consistency ratio is 0.001, which passes the consistency check.

[0116] In specific implementation, as a preferred embodiment of the present invention, such as Figure 3 As shown, step S8 includes: S81. Calculate the collision risk assessment item based on the overall ship risk after adopting candidate maneuvers. The formula is as follows:

[0117] S82. Calculate the distance loss item to reflect the cost of detour. The formula is as follows:

[0118] S83. Calculate the steering amplitude term used to reflect the steering amplitude. The formula is as follows:

[0119] S84. Calculate the avoidance liability assessment item that includes the International Maritime Collision Prevention Regulations and penalties for violations of the give-way duty. The formula is as follows:

[0120] Among them, when the candidate maneuver simultaneously meets the requirements of the International Maritime Collision Prevention Regulations and the obligation to avoid collision. If the value is 0, otherwise a positive penalty value is applied. ; S85. Calculate the safety distance evaluation term for candidate schemes whose closest distance is less than the ship-to-ship safety distance threshold. The formula is as follows:

[0121] in, For safe distance threshold, ; S86. The multi-stage cooperative collision avoidance problem is represented as a multi-objective constrained optimization problem. The five evaluation indicators are unified as collision risk, range loss, turning angle, avoidance responsibility, and safe distance. The comprehensive fitness function is then constructed as follows:

[0122] in, , , , , All are weighting coefficients; S87. The input to the improved differential evolution algorithm includes the decision variable boundaries, fitness function, and constraints of the vessel to be decided. The output is the vessel's turning amplitude, collision avoidance start time, and rerouting time. A chaotic mapping function is used. The initial individuals are generated within the boundaries of each decision variable, as shown in the following formula:

[0123]

[0124]

[0125] in, For chaotic parameters, the generated initial individuals cover the full range of decision variables; S88. Divide the overall population into Each subpopulation performs differential mutation in parallel within the same generation; individuals in each generation are sorted according to their fitness function values, and superior individuals are written into the elite archive and allowed to be used as mutation basis vectors or to replace inferior individuals; individuals that are consistently in the inferior range are eliminated or chaotically reinitialized to maintain population diversity. S89. Perform crossover and greedy selection, retaining experimental individuals with fitness no worse than their parents. The corresponding population partitioning, mutation, crossover, and selection relationships are as follows:

[0126]

[0127] The above formula employs a general bounded update framework for success parameter statistics. and Each by the first The scaling factor and crossover factor are calculated to generate better experimental individuals. and For update rate, The results are restricted to a given range; in this embodiment, sensitivity analysis confirms... Take 0.4 to 1.2. It exhibits good robustness when the value is between 0.3 and 0.7.

[0128] S810. Execute decisions ship by ship according to the determined cooperative collision avoidance priority: For the current critical ship, use the constructed comprehensive fitness function as the objective to solve for the turning amplitude, collision avoidance start time, and return-to-course time, while maintaining the current navigation state of the remaining undecided ships. The decision vector for each critical ship is defined as:

[0129] in, For the steering range, To avoid collision at the start time, The time for resumption of service; S811. After writing the obtained actions into the scene state, recalculate the relative motion parameters of all ship pairs and the safe distance threshold between ship pairs and update the encounter state. S812. Solve for the next critical vessel. Repeat steps S810 to S81 until the collaborative collision avoidance decision for all critical vessels is completed. Output the turning amplitude, collision avoidance start time, rerouting time and reference trajectory of each vessel.

[0130] In this embodiment, collaborative solving is performed ship-by-ship according to risk priority. After obtaining the decision vector for the current critical vessel, its turning and timing actions are written into the encounter state, and the DCPA, TCPA, and [other parameters] of all vessel pairs are recalculated. Then, the next key vessel is optimized. This avoids new local conflicts caused by multiple vessels making independent decisions at the same time.

[0131] like Figure 4 As shown, this embodiment of the invention also provides a multi-ship cooperative collision avoidance decision-making system considering risk coupling, implemented based on the above-mentioned multi-ship cooperative collision avoidance decision-making method considering risk coupling. The system includes: a state acquisition module, a relative motion parameter calculation module, a risk fusion module, a risk distance matrix construction module, a risk cluster identification module, a collision avoidance priority determination module, a weight determination module, and a cooperative optimization and decision output module, wherein: The status acquisition module is used to acquire the position, speed, heading, length and safety parameters of each ship in a multi-ship encounter scenario. The relative motion parameter calculation module is used to calculate the relative motion parameters between each pair of ships and the safe distance threshold between the pairs of ships based on the obtained positions, speeds and headings of each ship. The risk fusion module is used to calculate the objective collision risk based on relative motion parameters, determine the ship domain constraints by combining the ship pair safety distance threshold, and use prospect theory to perform nonlinear correction on the objective collision risk to construct a subjective-objective fusion risk value. The risk distance matrix construction module is used to symmetrize, normalize, and perform negative logarithmic mapping on the constructed subjective and objective integrated risk values ​​to construct a risk distance matrix. The risk cluster identification module is used to input the constructed risk distance matrix into the OPTICS clustering algorithm, generate an ordered ship sequence and reachability sequence through core distance and reachability distance, and extract continuous valley areas as local risk clusters based on the reachability distance threshold; The collision avoidance priority determination module is used to determine the processing order of risk clusters and the processing order of key ships within clusters based on the extracted cluster-level average risk of each local risk cluster, the maximum ship-to-ship risk within the cluster, and the global risk of each ship, so as to obtain the collaborative collision avoidance priority. The weight determination module is used to construct a judgment matrix with collision risk, range loss, turning radius, avoidance responsibility and safe distance as evaluation indicators, and to determine the weight of each evaluation indicator using the analytic hierarchy process. The collaborative optimization and decision output module is used to construct a comprehensive fitness function by weighting each evaluation index according to the collaborative collision avoidance priority and a determined weight. It uses an improved differential evolution algorithm to solve the turning amplitude, collision avoidance start time and rerouting time of each ship to be decided in turn. After each ship's action is determined, it updates the multi-ship encounter state and outputs the multi-ship collaborative collision avoidance decision scheme.

[0132] In this embodiment, the system can be deployed on an intelligent ship's onboard computer, a shore-based traffic service system, or a ship-shore collaborative computing platform. The processor executes the program in the memory, exchanges data between modules through the shipboard communication network, and sends the generated collision avoidance actions to the navigation control system; for manned vessels, the actions can also be displayed as driving assistance suggestions.

[0133] The embodiments of the present invention are described simply because they correspond to those in the embodiments above. For any similarities, please refer to the descriptions in the embodiments above, which will not be elaborated here.

[0134] Example 1: Identification of Risk Clusters of Twenty Ships This embodiment uses a two-dimensional open water scenario to verify the risk coupling relationship and the OPTICS risk cluster identification process. The initial states of the 20 vessels in the scenario are shown in Table 2. The simulation step size is 30 s, and the total number of simulation steps is 100, i.e., the total duration is 3000 s; the risk structure analysis times are the initial moment, step 30, step 60, and step 90. The OPTICS parameters are set as follows: Reachable distance extraction threshold ; Safety distance reference value nautical miles, safe time reference value Distance loss weight Time loss weight .

[0135] Table 2 Initial State of a Complex Encounter Scenario with Twenty Ships

[0136] In each simulation step, the objective collision risk and prospect theory fusion risk of each ship pair are updated, and then symmetry, normalization, and negative logarithmic mapping are performed to gradually update the risk distance matrix, which is used as the pre-calculated distance input for OPTICS. As shown in Figure 5, 5(a) corresponds to the initial moment, when the 20 ships are relatively dispersed in space and have not yet formed a stable local risk cluster; 5(b) corresponds to step 30, when local risk coupling relationships begin to appear between some ships; 5(c) corresponds to step 60, when the local risk relationships are further strengthened and gradually form multiple small-scale risk clusters; 5(d) corresponds to step 90, when the risk structures of C1, C2, and C3 tend to stabilize, indicating that the risk clusters evolve dynamically with the changes in ship positions and relative motion states.

[0137] like Figure 6 As shown, Figure 6 (a) and Figure 6 (b) Provide the risk matrices for step 90 with and without prospect theory. Figure 6 (c) and Figure 6(d) The corresponding OPTICS reachability threshold segmentation results are given. With the introduction of prospect theory, high-risk relationships are more concentrated in the risk matrix, and the boundaries of local high-risk areas are clearer; the corresponding reachability curves form more continuous valleys below the threshold, which is beneficial for OPTICS to extract stable local risk clusters. Without prospect theory, high-risk relationships are more dispersed, the reachability curves fluctuate greatly, and the valley continuity is weak; some potentially high-risk vessels are more easily identified as noise.

[0138] like Figure 7 As shown, Figure 7 (a) shows the final risk cluster identification result after introducing prospect theory in step 90. Figure 7 (b) shows the final risk cluster identification results in step 90 without the introduction of prospect theory. With the introduction of prospect theory, the number of ships covered by the risk clusters increases, and the local risk structures of C1, C2, and C3 become more complete. Without prospect theory, the risk coupling relationship is relatively weak, and some potentially risky ships still appear as noise points. Cluster division under both conditions directly adopts the corresponding risk structure from step 90, without recalculating or changing the original clustering results.

[0139] Table 3 Risk clustering results before and after introducing prospect theory

[0140] As shown in Table 3, after introducing prospect theory, the coverage rate increased from 0.40 to 0.65, the high-risk recall rate increased from 0.80 to 1.00, and the number of noisy vessels decreased from 12 to 7. Although the DBI changed from 0.21 to 0.27, the coverage penalty DBI decreased from 0.53 to 0.41, indicating that while maintaining the cluster compactness and separation evaluation, more relevant risky vessels were included in the effective risk cluster, and the coverage capability of risk structure identification and the ability to capture high-risk targets were improved.

[0141] like Figure 8 As shown, Figure 8 (a) The cluster-level risk distribution of each risk cluster is given under the condition of introducing prospect theory in step 90. Among them, C3 has the highest overall risk, while C1 and C2 have relatively low risk levels, thus forming the priority among risk clusters; Figure 8 (b) The global risk distribution of each vessel under the same conditions is given, with TS3, TS14, and TS18 having higher global risks. First, C1, C2, and C3 are sorted according to cluster-level risks, and then key vessels are identified within the clusters and globally according to the global risks of each vessel, which can form the risk priority required for subsequent collaborative collision avoidance decisions.

[0142] Example 2: Collision Avoidance by Ten Ships This embodiment uses a ten-ship encounter scenario as shown in Figure 9 to verify the complete process from risk cluster identification and priority determination to cooperative collision avoidance trajectory generation. The simulation step size is 20 seconds; the improved differential evolution algorithm has a maximum of 100 iterations, a population size of 50, and a decision variable dimension of [missing information]. The scaling factor ranges from 0.4 to 1.2, and the cross factor ranges from 0.3 to 0.7. Safety distance constraints use ship-to-ship thresholds. The initial states of each vessel are shown in Table 4. In the table, 0.9, 1.0, and 1.1 nautical miles are the safe radius values ​​obtained from the embodiment based on the vessel dimensions, safety dimension coefficients, or preset safety field parameters.

[0143] Table 4 Initial State of a Ten-Ship Collision Avoidance Scenario

[0144] The weights of the five evaluation items obtained through AHP—collision risk, range loss, steering angle, avoidance responsibility, and safe distance—are shown in Table 5. The maximum eigenvalue is 5.004, and the consistency ratio is 0.001, thus passing the consistency test.

[0145] Table 5 Weights of AHP Evaluation Items

[0146] Using the same scenario and constraints, the improved differential evolution algorithm is compared with the standard differential evolution algorithm, genetic algorithm, and... The algorithm was statistically compared, and the results are shown in Table 6.

[0147] Table 6 Statistical Comparison of Different Algorithms

[0148] The risk cluster identification results divided the ten ships into three effective risk clusters: C1, C2, and C3, with TS3 being a noisy ship. The key ship sequences obtained according to cluster-level risk and global ship risk are OS, TS4, TS6, TS2, and TS9. An improved differential evolution solution was then applied to these key ships sequentially, yielding the actions shown in Table 7.

[0149] Table 7 Key Vessel Phased Coordinated Collision Avoidance Actions

[0150] like Figure 10 As shown, the key vessels do not maneuver simultaneously, but rather initiate collision avoidance sequentially in steps 30, 32, 34, 38, and 40 according to risk priority, and gradually resume navigation after the risk has passed. This method ensures that subsequent vessels' decisions are based on the updated encounter status. The trajectory and dynamic interaction results show that the collision avoidance actions are continuous and have clear intent; 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-ship cooperative collision avoidance decision-making method considering risk coupling, characterized in that, include: S1. Obtain the position, speed, heading, length, and safety parameters of each ship in a multi-ship encounter scenario. S2. Based on the obtained positions, speeds and headings of each vessel, calculate the relative motion parameters between each pair of vessels and the safe distance threshold between the pairs of vessels. S3. Calculate the objective collision risk based on relative motion parameters, determine the ship domain constraints by combining the ship pair safety distance threshold, and use prospect theory to perform nonlinear correction on the objective collision risk to construct a subjective-objective integrated risk value. S4. Symmetricize, normalize, and perform negative logarithmic mapping on the constructed subjective and objective integrated risk values ​​to construct a risk distance matrix; S5. Input the constructed risk distance matrix into the OPTICS clustering algorithm to generate an ordered ship sequence and reachability sequence through core distance and reachability distance. Extract continuous valley areas as local risk clusters based on the reachability distance threshold. S6. Based on the average risk at the cluster level, the maximum ship-to-ship risk within the cluster, and the global risk of each ship, determine the processing order of risk clusters and the processing order of key ships within the cluster to obtain the collaborative collision avoidance priority. S7. Construct a judgment matrix using collision risk, range loss, turning radius, avoidance responsibility, and safe distance as evaluation indicators, and use the analytic hierarchy process (AHP) to determine the weight of each evaluation indicator. S8. Based on the priority of collaborative collision avoidance, a comprehensive fitness function is constructed by weighting each evaluation index according to the determined weights. An improved differential evolution algorithm is used to solve the turning amplitude, collision avoidance start time and rerouting time of each ship to be decided in turn. After each ship's action is determined, the multi-ship encounter state is updated, and the multi-ship collaborative collision avoidance decision scheme is output.

2. The multi-ship cooperative collision avoidance decision-making method considering risk coupling according to claim 1, characterized in that, Step S1 includes: S11. Receive multi-source data provided by automatic identification systems, radar, satellite positioning equipment and electronic chart systems; S12. Clean the multi-source data to form the time-lapse data for each ship. The state vector is as follows: in, For position coordinates, , For speed, , For the course, As captain, For the radius of the safety zone.

3. The multi-ship cooperative collision avoidance decision-making method considering risk coupling according to claim 1, characterized in that, Step S2 includes: S21. For any ship With ships The relative position vectors are constructed based on the obtained positions of each ship, as shown in the following formula: in, Indicates a ship Location, Indicates a ship Location; S22. For any ship With ships The relative velocity vector is constructed based on the obtained speed and heading of each ship, as shown in the following formula: in, Indicates a ship speed, Indicates a ship The speed of the ship; S23, Calculate real-time distance Relative velocity Relative orientation and speed ratio The formula is as follows: in, Used to map angles to a preset angle range. To prevent positive numbers with a denominator of zero; S24. Calculate the minimum meeting time and nearest meeting distance based on the constructed relative position vector and relative velocity vector, using the following formula: S25. Calculate the safety zone radius of each vessel based on the obtained ship length and safety zone parameters. And calculate the ship With ships The safe distance threshold between them is calculated using the following formula: in, For safety scale coefficient, Here is the conversion factor between meters and nautical miles; S26. Identify the encounter, crossing, or overtaking relationship based on the calculated relative bearing and heading difference, and determine the yielding responsibility and permissible turning direction in accordance with the International Regulations for Preventing Collisions at Sea.

4. The multi-ship cooperative collision avoidance decision-making method considering risk coupling according to claim 1, characterized in that, Step S3 includes: S31. Based on the calculated nearest encounter distance (DCPA), minimum encounter time (TCPA), real-time distance, relative orientation, and velocity ratio, calculate the objective collision risk using the following formula: in, , , , and The weights are non-negative, and the sum of the weights is 1. , , , and For the corresponding normalized risk factor; S32. Determine the vessel domain constraints based on the safety domain radius and the obtained location, and establish a circular vessel domain around each vessel, as shown in the following formula: S33. Employing prospect theory to perform nonlinear correction on objective collision risk, constructing a combined subjective and objective risk value, specifically including: The objective collision risk level is mapped to a probabilistic surrogate quantity to avoid numerical anomalies in the probability weighting function when the probability is 0 or 1: The nonlinear perception of low-probability risks is described using a probability weighting function from prospect theory: in, and These are probability weight parameters. This can enhance the perceived weight of low-probability risks; The distance loss and time loss are calculated based on the ship-to-ship safe distance threshold, using the following formula: in, and These are reference values ​​for safe distance and safe time, respectively. With safe distance threshold Maintain consistency; The severity of the risk is obtained by weighting the distance loss and time loss, as shown in the following formula: in, and All are weighting coefficients, and ; Considering loss aversion and diminishing marginal sensitivity, a subjective value and a risk value combining subjective and objective factors are constructed, as follows: in, For the ship Subjective value, The loss aversion coefficient, Assess the severity of the risk. The sensitivity coefficient, Combine subjective and objective risk values. This is the probabilistic proxy quantity after mapping objective collision risk. The scale parameter of the Prelec function. This is the curvature parameter of the Prelec function.

5. A multi-ship cooperative collision avoidance decision-making method considering risk coupling according to claim 1, characterized in that, Step S4 includes: S41. Symmetrically transform the constructed subjective-objective fusion risk value to obtain the symmetrical subjective-objective fusion risk, as shown in the following formula: S42. Normalize using the maximum value of the non-diagonal elements within the scene to form... : S43. Construct the risk distance using a negative logarithmic mapping, as shown in the following formula: in, To prevent positive numbers from having zero values ​​during normalization and logarithmic operations, the diagonal elements are fixed to zero, thus obtaining the risk distance matrix.

6. A multi-ship cooperative collision avoidance decision-making method considering risk coupling according to claim 1, characterized in that, Step S5 includes: S51, the negative logarithmic risk distance matrix As a pre-computed distance input for the OPTICS clustering algorithm; S52, For ships The core distance is the risk distance to the MinPts-th nearest neighbor, and the formula is as follows: S53, Regarding current vessels and its neighboring vessels The reachable distance is calculated using the following formula: S54 and OPTICS continuously update the reachability distances of each vessel, resulting in an ordered vessel sequence. and the corresponding reachability sequence ; S55. In the reachability distance curve, continuous valleys correspond to locally high-risk clusters, and peaks correspond to the separation between risk clusters; let the clustering distance threshold be... Extracting continuous satisfaction Sequence segments with at least MinPts of ships are considered local risk clusters. ; S56. The effectiveness of risk cluster identification is evaluated using coverage, Davies-Bouldin index (DBI), coverage penalty (DBI), and high-risk recall, as shown in the following formula: Among them, the higher the coverage and high-risk recall, and the lower the DBI and coverage penalty DBI, the more reliable the local risk cluster identification results are.

7. A multi-ship cooperative collision avoidance decision-making method considering risk coupling according to claim 1, characterized in that, Step S6 includes: S61. For localized risk clusters The cluster-level average risk and the maximum ship pair risk within the cluster are calculated using the following formulas: S62, Calculation of Ships The global risk in the entire multi-ship risk network is calculated as follows: S63, First, calculate the average risk at the cluster level. Risk clusters are sorted in descending order. When the average risks of two risk clusters are close, the cluster with the highest ship-to-ship risk is selected. Prioritize larger risk clusters and determine the processing order for risk clusters; S64. For vessels within the same risk cluster, determine the key vessel sequence according to the global risk descending order, and combine this with encounter responsibility to exclude straight-running vessels that do not require active maneuvering, to obtain the cooperative collision avoidance priority.

8. A multi-ship cooperative collision avoidance decision-making method considering risk coupling according to claim 1, characterized in that, Step S7 includes: S71. Construct a five-order judgment matrix using collision risk, range loss, turning angle, avoidance responsibility, and safe distance as evaluation indicators. ; S72, Solving the judgment matrix The largest eigenvalue and its corresponding eigenvector are obtained, and the eigenvector is normalized to obtain the weight vector. ; S73. Calculate the consistency ratio and perform a consistency test. The formula is as follows: in, As the average random consistency index, when If the value is less than 0.1, the weight vector is accepted; otherwise, the judgment matrix is ​​adjusted and the calculation is repeated.

9. A multi-ship cooperative collision avoidance decision-making method considering risk coupling according to claim 1, characterized in that, Step S8 includes: S81. Calculate the collision risk assessment item based on the overall ship risk after adopting candidate maneuvers. The formula is as follows: S82. Calculate the distance loss item to reflect the cost of detour. The formula is as follows: S83. Calculate the steering amplitude term used to reflect the steering amplitude. The formula is as follows: S84. Calculate the avoidance liability assessment item that includes the International Maritime Collision Prevention Regulations and penalties for violations of the give-way duty. The formula is as follows: Among them, when the candidate maneuver simultaneously meets the requirements of the International Maritime Collision Prevention Regulations and the obligation to avoid collision. If the value is 0, otherwise a positive penalty value is applied. ; S85. Calculate the safety distance evaluation term for candidate schemes whose closest distance is less than the ship-to-ship safety distance threshold. The formula is as follows: in, For safe distance threshold, ; S86. The multi-stage cooperative collision avoidance problem is represented as a multi-objective constrained optimization problem. The five evaluation indicators are unified as collision risk, range loss, turning angle, avoidance responsibility, and safe distance. The comprehensive fitness function is then constructed as follows: in, , , , , All are weighting coefficients; S87, Using a chaotic mapping function The initial individuals are generated within the boundaries of each decision variable, as shown in the following formula: in, For chaotic parameters, the generated initial individuals cover the full range of decision variables; S88. Divide the overall population into Each subpopulation performs differential mutation in parallel within the same generation; individuals in each generation are sorted according to their fitness function values, and superior individuals are written into the elite archive and allowed to be used as mutation basis vectors or to replace inferior individuals; individuals that are consistently in the inferior range are eliminated or chaotically reinitialized to maintain population diversity. S89. Perform crossover and greedy selection, retaining experimental individuals with fitness no worse than their parents. The corresponding population partitioning, mutation, crossover, and selection relationships are as follows: The above formula employs a general bounded update framework for success parameter statistics. and Each by the first The scaling factor and crossover factor are calculated to generate better experimental individuals. and For update rate, Limit the results to a given range; S810. Execute decisions ship by ship according to the determined cooperative collision avoidance priority: For the current critical ship, use the constructed comprehensive fitness function as the objective to solve for the turning amplitude, collision avoidance start time, and return-to-course time, while maintaining the current navigation state of the remaining undecided ships. The decision vector for each critical ship is defined as: in, For the steering range, To avoid collision at the start time, The time for resumption of service; S811. After writing the obtained actions into the scene state, recalculate the relative motion parameters of all ship pairs and the safe distance threshold between ship pairs and update the encounter state. S812. Solve for the next critical vessel. Repeat steps S810 to S81 until the collaborative collision avoidance decision for all critical vessels is completed. Output the turning amplitude, collision avoidance start time, rerouting time and reference trajectory of each vessel.

10. A multi-ship cooperative collision avoidance decision-making system considering risk coupling, implemented based on the multi-ship cooperative collision avoidance decision-making method considering risk coupling as described in any one of claims 1-9, characterized in that, include: The module comprises a state acquisition module, a relative motion parameter calculation module, a risk fusion module, a risk distance matrix construction module, a risk cluster identification module, a collision avoidance priority determination module, a weight determination module, and a collaborative optimization and decision output module, wherein: The status acquisition module is used to acquire the position, speed, heading, length and safety parameters of each ship in a multi-ship encounter scenario. The relative motion parameter calculation module is used to calculate the relative motion parameters between each pair of ships and the safe distance threshold between the pairs of ships based on the obtained positions, speeds and headings of each ship. The risk fusion module is used to calculate the objective collision risk based on relative motion parameters, determine the ship domain constraints by combining the ship pair safety distance threshold, and use prospect theory to perform nonlinear correction on the objective collision risk to construct a subjective-objective fusion risk value. The risk distance matrix construction module is used to symmetrize, normalize, and perform negative logarithmic mapping on the constructed subjective and objective integrated risk values ​​to construct a risk distance matrix. The risk cluster identification module is used to input the constructed risk distance matrix into the OPTICS clustering algorithm, generate an ordered ship sequence and reachability sequence through core distance and reachability distance, and extract continuous valley areas as local risk clusters based on the reachability distance threshold; The collision avoidance priority determination module is used to determine the processing order of risk clusters and the processing order of key ships within clusters based on the extracted cluster-level average risk of each local risk cluster, the maximum ship-to-ship risk within the cluster, and the global risk of each ship, so as to obtain the collaborative collision avoidance priority. The weight determination module is used to construct a judgment matrix with collision risk, range loss, turning radius, avoidance responsibility and safe distance as evaluation indicators, and to determine the weight of each evaluation indicator using the analytic hierarchy process. The collaborative optimization and decision output module is used to construct a comprehensive fitness function by weighting each evaluation index according to the collaborative collision avoidance priority and a determined weight. It uses an improved differential evolution algorithm to solve the turning amplitude, collision avoidance start time and rerouting time of each ship to be decided in turn. After each ship's action is determined, it updates the multi-ship encounter state and outputs the multi-ship collaborative collision avoidance decision scheme.

Images