Ship collaborative collision avoidance decision-making method based on multilayer coding genetic algorithm

Abstract

The invention discloses a ship collaborative collision avoidance decision-making method based on a multilayer coding genetic algorithm. The method comprises the following steps: obtaining and processing related information of each ship in an encounter situation based on a sensor; based on a collision risk degree model, constructing a dynamic early warning mechanism in a multi-ship encounter scene: constructing a multi-stage ship collision avoidance strategy based on a queuing theory theoretical framework, establishing a multi-ship collision avoidance decision queuing system with specific space-time coupling characteristics, introducing a dynamic time window distribution mechanism and a space-time constraint coupling network modeling method, and establishing a multi-ship collision avoidance decision queuing system; multi-objective optimization is carried out on the ship decision scheme of the first priority gradient, and a ship intelligent collision avoidance strategy with dynamic adaptability is provided; a multi-ship collaborative collision avoidance decision optimization model is constructed on the basis of a genetic evolution mechanism, a multi-target adaptation function is integrated, a decision scheme when multiple groups of ships meet in a first priority is determined through the synergistic effect of dynamic priority management and multi-layer coding optimization, and hierarchical analysis of decisions in a multi-ship meeting situation is achieved.

Description

Technical Field

[0001] This invention belongs to the field of autonomous collision avoidance decision-making for ships, and particularly relates to a collaborative collision avoidance decision-making method for ships based on a multi-layer coded genetic algorithm. Background Technology

[0002] Autonomous collision avoidance decision-making technology, as a core component of the autonomous surface vessel (MASS) intelligent navigation system, undertakes the critical task of generating safe navigation strategies in real time in complex marine environments. By deeply integrating the International Regulations for Preventing Collisions at Sea (ICMLS) with practical maritime wisdom, this technology can not only accurately avoid static obstacles such as islands, reefs, and shipwrecks, but also dynamically address collision risks in multi-vessel encounters. While improving navigation safety, it significantly optimizes energy efficiency, providing crucial technical support for the green transformation of the shipping industry.

[0003] The development of autonomous collision avoidance decision-making technology for ships is currently undergoing a profound methodological transformation. Traditional decision-making systems based on classical mathematical models have gradually evolved into innovative paradigms that integrate artificial intelligence, control theory, navigation technology, and other disciplines. This technological leap not only strengthens the ability to digitally interpret international rules but also constructs a cognitive framework that conforms to real-world navigation scenarios, propelling collision avoidance decision-making systems from theoretical optimization to practical applicability.

[0004] From an algorithmic architecture perspective, existing collision avoidance decision modeling methods can be divided into three main technical schools: the first is mathematical modeling based on operations research theory, which achieves path planning by constructing constrained optimization models; the second is artificial intelligence methods centered on machine learning, which rely on data-driven models to uncover navigation patterns; and the third is a deep integration of the above two, which achieves the organic unity of rule constraints and learning capabilities through a hybrid intelligent architecture. This technical classification system not only reflects the innovative characteristics of interdisciplinary collaboration but also provides a clear roadmap for subsequent technological evolution.

[0005] (1) Mathematical Model Algorithm

[0006] These algorithms typically represent the external environment and ship dynamics using a relatively accurate mathematical or physical model. They solve for the future objective state and optimal collision avoidance decision scheme of a single ship according to a strictly defined decision process. They are characterized by fast convergence speed and deterministic solution. These algorithms mainly include geometric analytical methods, model predictive control, speed barrier methods, artificial potential field methods, game theory, etc.

[0007] (2) Artificial intelligence algorithms and soft computing

[0008] The essence of this type of algorithm is to obtain a decision scheme that meets the design requirements within a given search space, based on indicators such as safety, efficiency, rule applicability, path smoothness, and yaw distance, by setting corresponding knowledge systems, mapping rules, and constraints under various constraints. This type of algorithm mainly includes knowledge systems, artificial neural networks, fuzzy logic, genetic algorithms, ant colony algorithms, particle swarm optimization algorithms, and multi-agent algorithms.

[0009] (3) Hybrid intelligent algorithm

[0010] Hybrid intelligent systems refer to intelligent systems that, in the process of solving complex real-world problems, utilize the differences and complementarities of various intelligent and non-intelligent technologies in terms of basic theory, supporting technologies, and application perspectives, and adopt different hybrid approaches to obtain intelligent systems with stronger knowledge expression and reasoning abilities, higher operating efficiency, and stronger problem-solving capabilities. These algorithms mainly include fuzzy neural systems, hybrid expert systems, and humanoid intelligent navigation systems.

[0011] The current state of research on ship collision avoidance decision-making technology in complex encounter situations reflects that existing technologies have failed to achieve the technological leap from single-ship intelligence to multi-ship collaboration, from theoretical safety to practical applicability, and from rule compliance to intelligent decision-making: (1) The failure to construct a dynamic priority assessment framework for multi-ship encounter situations leads to a lack of responsibility for collision avoidance; (2) The lack of global collaborative optimization capabilities causes the collision avoidance scheme to fall into a local optimum trap; (3) Existing cluster optimization-based models fail to quantify and analyze the International Maritime Collision Regulations and collision avoidance experience, thereby weakening the effectiveness and practicality of decision-making schemes to some extent. Summary of the Invention

[0012] To address the technical challenges of multi-ship cooperative collision avoidance in complex maritime traffic environments, this invention employs the following technical solution: a ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm, comprising the following steps:

[0013] S1: Acquire and process relevant information of each ship in the encounter situation based on sensors;

[0014] S2: Based on the collision hazard model, a dynamic early warning mechanism is constructed for multi-ship encounter scenarios: when the distance between any two ships triggers the hazard threshold, the spatial collision hazard and time urgency indicators are coupled to construct a continuous quantification priority gradient model based on the 0-1 scale. The priority gradient is generated through dynamic quantification of spatiotemporal urgency to achieve accurate clustering of collision risks and determination of collision avoidance responsibilities.

[0015] S3: Based on the queuing theory framework, a multi-stage ship collision avoidance strategy is constructed. Specifically, this includes designing a parallel service architecture for multi-ship collaborative optimization, establishing a multi-ship collision avoidance decision queuing system with specific spatiotemporal coupling characteristics, introducing a dynamic time window allocation mechanism and a spatiotemporal constraint coupling network modeling method, performing multi-objective optimization on the ship decision scheme with the first priority gradient, and developing a dynamically adaptive intelligent ship collision avoidance strategy.

[0016] S4: Based on the genetic evolution mechanism, a multi-ship collaborative collision avoidance decision optimization model is constructed. A multi-level coding genetic algorithm is used to integrate multi-objective adaptation functions. Through the synergistic effect of dynamic priority management and multi-level coding optimization, the decision schemes for multiple ships meeting in the first priority are determined, realizing hierarchical analysis of decision-making in multi-ship encounter situations.

[0017] Furthermore, the collision hazard model is constructed as follows:

[0018] The nearest encounter distance and the time to reach the nearest encounter distance were established using a circular ship domain model and a Gaussian fitting method. Spatial collision hazard and temporal collision hazard were established respectively, and their synthesis method was determined.

[0019] The determination of the spatial collision hazard level is as follows:

[0020] The risk level of a space collision between the two ships is set as R. s The membership function r is established based on the asymmetric Gaussian equation. s for:

[0021]

[0022] Where, take r0 = 0.5, when f s When (t) = 1, r s =0.5, which defines the spatial collision risk value when the minimum safe meeting distance between two ships is equal to the radius of the ship's territory;

[0023] The determination of the time-based collision risk level is as follows:

[0024] The time-collision risk set between this vessel and other vessels is R. t R is established through the asymmetric Gaussian equation. t Membership function r t for:

[0025]

[0026] Where, when f t When (t) = 1, define r t =0.5 is the critical value of the collision risk level at the time when the encounter situation is formed, that is, the time membership function value corresponding to the two ships approaching from the applicable distance to the safe encounter distance in a specific encounter situation;

[0027] The synthesis method is determined by synthesis rules. The synthesis rules use the method of synthetic space collision risk and time collision risk to represent the actual collision risk between ships. is a synthesis operator:

[0028]

[0029] ① When r s < r0, it is considered that there is no collision risk between ships and no action needs to be taken;

[0030] ② When r s ≥ r0, r t < r0, it is considered that there is no collision risk between ships and no action needs to be taken;

[0031] ③ When r s ≥ r0, r t ≥ r0, there is a collision risk between ships, and the give-way ship needs to take avoidance action.

[0032] Furthermore: The coupling of space collision risk and time urgency index constructs a continuous quantization priority gradient model based on 0-1 scale. The process of generating priority gradient through spatio-temporal urgency dynamic quantization and realizing accurate clustering of collision risks and definition of collision avoidance responsibilities is as follows:

[0033] S21: Use the on-board navigation system to obtain the dynamic and static data of each ship;

[0034] S22: Initialize the main priority queue Q1 and the secondary priority queue Q2 as empty sets;

[0035] S23: Initialize the index parameter k = 1 and construct the pairing pointer q = k + 1;

[0036] S24: Calculate the spatio-temporal composite risk index of the current ship i = k and the candidate ship j = q: spatial collision risk SCR ij and time collision risk TCR ij ;

[0037] S25: When SCR ij ∈ [r0 + Δr, 1] and TCR ij ∈ [r0 + Δr, 1] hold simultaneously, classify the ship pair [i, j] into the Q1 queue, otherwise, go to step S26;

[0038] S26: Increment the pairing pointer q = q + 1;

[0039] S27 If q≤N, continue to step (4); otherwise check the state of index k; if k≥N, construct Q2=P-Q1, otherwise increment k=k+1 and reset q=k+1 and return to step (3);

[0040] (8) When the full ship traversal is completed and k≥N, a secondary priority queue Q2=P-Q1 is generated through international verification operation;

[0041] (9) Finally, output the classification results of the dual priority queues Q1 (high priority) and Q2 (second priority).

[0042] Furthermore: the establishment of a multi-ship collision avoidance decision queuing system with specific spatiotemporal coupling characteristics includes:

[0043] Input process modeling module: used to determine whether a ship arrives alone or in a convoy through spatiotemporal correlation analysis;

[0044] An improved Poisson process is used to model the ship arrival pattern, combined with a dual-threshold triggering mechanism for collision risk. When the real-time monitoring value exceeds the preset threshold, the queuing service process is automatically initiated to confirm the number and status of arriving ships in the first priority, ensuring timely response to potential collision risks.

[0045] Queuing and Service Mechanism Module: Includes

[0046] Queue capacity configuration submodule: Used to set the traffic carrying capacity of a limited buffer zone to simulate the actual waterway. When the number of ships exceeds the capacity threshold, external flow control measures will be triggered to prevent traffic congestion and collision accidents.

[0047] The hierarchical priority rule submodule is used to implement a dual-priority scheduling strategy. Service agencies focus on handling high-urgency vessels in the Q1 queue, adopting a preemptive service model to ensure that critical decisions can be executed first, thereby effectively reducing the risk of collision.

[0048] Service Desk Configuration and Service Module: Based on the characteristics of multiple ship encounters, a collaborative optimization model is established using a multi-layer coded genetic algorithm, and a single service desk is used to perform simultaneous decision optimization for ships with the same priority.

[0049] Furthermore: the multi-layer coding genetic algorithm adopts a real-number coding scheme, expanding the single-layer chromosome into a multi-layer structure. Each layer of coding corresponds to the collision avoidance decision scheme of a single ship in the first priority queue, realizing the global expression of multi-ship cooperative strategies by a single chromosome, wherein:

[0050] The individual coding strategy is as follows:

[0051] Floating-point encoding is used to encode the turning angle φi and turning operation time ti for each ship; the number of chromosome layers is dynamically determined by the number of ships with the highest priority, and each layer corresponds to a single ship's decision-making scheme, ensuring that:

[0052] The original course is automatically restored after a turning maneuver.

[0053] Decision variables are directly related to the kinematic properties of the ship;

[0054] The population initialization rule has the following constraints during the initialization phase:

[0055] The turning angle range for yielding vessels is: π / 12 < |φ i |<π / 3;

[0056] Turning time limit: 300s < t i <1800s;

[0057] Furthermore, the multi-objective adaptation function is constructed based on distance, safety, path smoothness, and rerouting indicators, as detailed below:

[0058]

[0059] In the formula: D e Minimum meeting distance between ships; D s : Radius of the security domain;

[0060] Ship trajectory;

[0061] τ: Path smoothness weighting coefficient;

[0062] N: Number of ships with first priority.

[0063] Furthermore: the genetic operation operators of the multi-layer encoded genetic algorithm include selection operation operators;

[0064] The operation flow of the selection operator includes the following core steps:

[0065] S411: Determine the probability allocation mechanism;

[0066] First, calculate the probability of each individual in the population being selected in the offspring population.

[0067]

[0068] In the formula, F i k Let N be the fitness value of the i-th individual in population k, and N be the population size. M Population size;

[0069] S412: Conduct individual screening;

[0070] Generate a random number P1 between [0,1], and find a number that satisfies P1. <F i k The required chromosomes, rotate N M The next roulette wheel yields a new generation of population with a size of N;

[0071] S413: Determine the elite retention strategy;

[0072] The top 5% of individuals in terms of fitness are directly admitted to the next generation, and the remaining spots are determined through a roulette-like competition mechanism.

[0073] Furthermore: the genetic operation operator of the multi-layer coding genetic algorithm includes a crossover operation operator, and the specific execution steps of the crossover operator are as follows:

[0074] S421: Randomly select two chromosomes from the population and decide whether to perform the crossover operation based on the crossover probability;

[0075] S422: Randomly select the crossover point for individuals and check the feasibility of the two new individuals generated. If the requirements are not met, the crossover operation is repeated.

[0076] S423: Repeat the above steps N times to complete the crossover operation for a single population.

[0077] Furthermore: the genetic operation operator of the multi-layer coding genetic algorithm includes a mutation operation, and the execution steps of the mutation operation are as follows:

[0078] S431: Randomly select the target chromosome in the current population and determine whether to perform this round of gene modification operation based on the preset mutation probability pm.

[0079] S432: After randomly determining the mutation sites, use a mean of μ and a variance of P. 2 The Gaussian perturbation strategy generates new gene values, and local neighborhood search is implemented to enhance the algorithm's exploration capability, while retaining the original gene positions generated by heuristic initialization;

[0080] S433: Repeat the above operation N times to achieve population variation and renewal through gradual gene perturbation.

[0081] This invention provides a ship collaborative collision avoidance decision-making method based on a multi-layer coded genetic algorithm, which constructs a spatiotemporally coupled risk assessment system, breaks through the traditional passive response mode, realizes the decision-making paradigm shift from "post-event handling" to "pre-event prediction", and significantly improves the risk response capability in complex waters with multiple ship interactions.

[0082] By reconstructing the decision-making process through queuing theory, unstructured collision avoidance problems are transformed into standardized service processes, establishing a closed-loop management system of "perception-assessment-decision-execution," which greatly improves the systematic nature and execution efficiency of multi-ship collaborative decision-making.

[0083] A multi-layer coding collaborative evolution mechanism is proposed, and an innovative algorithm architecture design is adopted to achieve a dynamic balance between collective intelligence and individual rationality. While ensuring the compliance of decision-making, the global optimization capability in multi-ship encounter scenarios is significantly improved.

[0084] This invention proposes an intelligent decision-making method based on a multi-layered coded genetic algorithm. By constructing a spatiotemporally coupled priority quantification model and a multi-stage collaborative optimization framework, it effectively addresses the core shortcomings of traditional technologies, such as ambiguous division of avoidance responsibilities, conflicting decision-making sequences, and insufficient collaborative efficiency. This method innovatively integrates spatial collision risk and time urgency indicators, establishing a 0-1 continuously quantified priority gradient model to achieve a paradigm shift from single-ship obstacle avoidance to group collaborative decision-making. Based on queuing theory, a three-level parallel service architecture is designed, and through dynamic time window allocation and a spatiotemporal constraint network, spatiotemporal decoupling of multi-ship avoidance trajectories is achieved. A multi-population genetic algorithm with a three-dimensional coded structure is employed, integrating multi-objective adaptation functions such as rule compliance, navigation economy, and safety margin to achieve global optimization of the decision scheme. Attached Figure Description

[0085] 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.

[0086] Figure 1 It is a four-stage collision avoidance model;

[0087] Figure 2 It is a priority gradient calculation model;

[0088] Figure 3 It is a collision avoidance decision queuing model;

[0089] Figure 4 It is the design of multi-stage collision avoidance strategies for ships;

[0090] Figure 5 It is the standard genetic algorithm process;

[0091] Figure 6 This is a diagram illustrating a ship's turning to avoid a collision.

[0092] Figure 7 It is a multi-layered encoded genetic algorithm technique;

[0093] Figure 8 This is a diagram of a single-point intersection.

[0094] Figure 9 (a) Initial spatial position of Scenario 1, (b) Optimized path of Scenario 1, (c) Distance change curve of Scenario 1, (d) Rudder angle and speed change curve of Scenario 1;

[0095] Figure 10 (a) Initial spatial position in scenario 2, (b) Optimized path in scenario 2, (c) Distance change curve in scenario 2, (d) Rudder angle and speed change curve in scenario 2. Detailed Implementation

[0096] It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0097] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0098] A ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm includes the following steps:

[0099] S1: Collect and process relevant information of each ship in the encounter situation based on sensors;

[0100] S2: Based on the collision hazard model, a dynamic early warning mechanism is constructed for multi-ship encounter scenarios: when the distance between any two ships triggers the hazard threshold, the spatial collision hazard and time urgency indicators are coupled to construct a continuous quantification priority gradient model based on the 0-1 scale. The priority gradient is generated through dynamic quantification of spatiotemporal urgency to achieve accurate clustering of collision risks and determination of collision avoidance responsibilities.

[0101] S3: Based on the queuing theory framework, a multi-stage ship collision avoidance strategy is constructed. Specifically, this includes designing a parallel service architecture for multi-ship collaborative optimization, establishing a multi-ship collision avoidance decision queuing system with specific spatiotemporal coupling characteristics, introducing a dynamic time window allocation mechanism and a spatiotemporal constraint coupling network modeling method, performing multi-objective optimization for the ship decision scheme with the first priority gradient, and developing a dynamically adaptive intelligent ship collision avoidance strategy.

[0102] S4: Based on the genetic evolution mechanism, a multi-ship collaborative collision avoidance decision optimization model is constructed. A multi-level coding genetic algorithm is adopted, and multi-objective adaptation functions are integrated. Through the synergistic effect of dynamic priority management and multi-level coding optimization, the decision schemes for multiple ships meeting in the first priority are determined, realizing the hierarchical analysis of decision-making in multi-ship encounter situations.

[0103] The steps S1 / S2 / S3 / S4 are executed sequentially;

[0104] The system collects information related to encountered vessels based on sensors, including heading, speed, and bearing. Then, collision avoidance parameters such as DCPA, TCPA, distance scale factor, and time scale factor are calculated from this information.

[0105] Figure 1 It is a four-stage collision avoidance model;

[0106] According to the framework of the International Regulations for Preventing Collisions at Sea (IRCS), "collision hazard" is a core legal concept, and its determination directly triggers the application of multiple clauses. Based on the dynamic evolution of collision hazards between vessels, the legal relationship for collision avoidance can be divided into four stages: freedom of action, formation of an imminent situation, critical danger, and the latest point of rudder application. The rights and obligations of the giving-way vessel and the vessel proceeding in the straight path differ significantly in each stage, reflected in the gradient changes in the priority and behavioral constraint intensity of special clauses such as Articles 12-17 of the Regulations under different stages of hazard.

[0107] ① Free Action Phase

[0108] At long distances, before an encounter situation is formed, the two ships are not bound by the encounter situation and are therefore free to take action.

[0109] ② Collision Hazard Stage

[0110] When two vessels approach each other to the point that the risk of collision reaches the threshold, and an initial encounter situation is formed, the vessel giving way should take significant collision avoidance actions as early as possible to ensure that the vessels pass each other at a safe distance, while the vessel traveling in the straight direction should maintain its course and speed accordingly.

[0111] ③ Urgent Situation Stage

[0112] When the giving vessel is clearly not taking appropriate action as required by the Rules, the vessel proceeding in the straight shall sound the whistle as prescribed in Rule 34, paragraph 4, and may take maneuvering action independently to avoid a collision. However, in a cross-traffic situation, the vessel proceeding in the straight shall generally not turn to port of the giving vessel on its port side.

[0113] ④ Imminent danger

[0114] Regardless of the cause, when the risk of collision between vessels reaches a certain level and the actions of the yielding vessel alone cannot prevent a collision, the vessel in the designated straight course should also take actions that are most conducive to avoiding a collision.

[0115] The collision hazard model is constructed as follows:

[0116] Ship collision hazard is considered a membership function of the set of ships that may collide with a target ship under certain navigation and conditions. Among these, the distance to the closest point of approach (DCPA) and time to the closest point of approach (TCPA) are considered the two most important factors influencing collision hazard. A circular ship domain model and Gaussian fitting method are used to establish spatial and temporal collision hazards, and their synthesis method is determined.

[0117] (1) Space collision risk

[0118] For any encounter situation between two ships, given the position vectors of the local and target ships, the distance scale factor between the two ships is defined as:

[0119] f s (t)=DCPA / D S (1)

[0120] Among them, f s (t) is the distance scale factor between the two ships, and Ds is the radius of the ship's domain.

[0121] The risk level of a space collision between the two ships is set as R. s The membership function r is established based on the asymmetric Gaussian equation. s for:

[0122]

[0123] Where, take r0 = 0.5, when f s When (t) = 1, r s =0.5, which defines the spatial collision risk value when the minimum safe encounter distance between two ships is equal to the radius of the ship's territory.

[0124] (2) Time Collision Risk

[0125] Similarly, for any two ships meeting, the time scale factor for the two ships is defined as:

[0126] f t (t) = TCPA / t f0 (3)

[0127]

[0128] Among them, f t (t) is the time scale factor of the two ships, is the time constant, Ra is the applicable distance for a specific encounter situation, and TCPA Ra is the TCPA corresponding to when the distance between the two ships is Ra; actually, it is very difficult to define a t f0 for all encounter situations with spatial collision risks. In the set collision avoidance decision-making procedure, the index is set as the smaller value of TCPA Ra and 20 minutes.

[0129] The time collision risk set of the own ship and the other ship is R t , and the membership function r t of R is established through the asymmetric Gaussian equation t as follows:

[0130]

[0131] Among them, when f t (t) = 1, r t is defined as 0.5, which is the critical value of the time collision risk corresponding to the formation of the encounter situation, that is, the time membership function value corresponding to the time when the two ships approach from the applicable distance of the specific encounter situation to the safe encounter distance.

[0132] (3) Composition rule

[0133] To reflect the psychological cognitive process of ship drivers regarding ship collision risks, the method of synthesizing spatial collision risk and time collision risk is used to represent the actual collision risk between ships, and is the composition operator.

[0134]

[0135] ① When r s < r0, it is considered that there is no collision risk between the ships and no action needs to be taken;

[0136] ② When r s ≥ r0, r t < r0, it is considered that there is no collision risk between the ships and no action needs to be taken;

[0137] ③ When r s ≥ r0, r t ≥ r0, there is a collision risk between the ships, and the give-way ship needs to take avoidance actions.

[0138] M ships sailing in a specific sea area form a traffic flow system, and its state vector is used to characterize the traffic safety situation of this water area. The specific representation form is as follows:

[0139]

[0140] In the formula, q represents the state vector of the system. i The state vector of ship i corresponds to the spatiotemporal correlation characteristics, and x represents the position vector of ship i.

[0141] To address the conflict in the allocation of collision avoidance responsibilities in complex encounter scenarios, a collision avoidance responsibility gradient partitioning model is introduced, and a hierarchical priority service scheduling mechanism is designed. This mechanism divides service objects into two responsibility gradients (e.g., ... Figure 2 As shown in the diagram, the service targets are expanded from individual ships to ship groups. The specific execution steps are as follows:

[0142] (1) Use the shipborne navigation system to obtain dynamic and static data of each ship;

[0143] (2) Initialize the primary priority queue Q1 and the secondary priority queue Q2 as empty sets;

[0144] (3) Initialize the index parameter k = 1, and construct the pairing pointer q = k + 1;

[0145] (4) Calculate the spatiotemporal composite risk index of the current ship i=k and the candidate ship j=q: Spatial Collision Risk Index (SCR) ij Time Collision Risk (TCR) ij ;

[0146] (5) When SCR is satisfied ij ∈[r0+Δr,1] and TCR ij If both ∈[r0+Δr,1] are true, the ship pair [i,j] is added to queue Q1; otherwise, proceed to step (6).

[0147] (6) Increment the paired pointer q = q + 1;

[0148] (7) If q≤N, continue to step (4); otherwise check the state of index k; if k≥N, construct Q2=P-Q1, otherwise increment k=k+1 and reset q=k+1 and return to step (3);

[0149] (8) When the full ship traversal is completed and k≥N, a secondary priority queue Q2=P-Q1 is generated through international verification operation;

[0150] (9) Finally, output the classification results of the dual priority queues Q1 (high priority) and Q2 (second priority).

[0151] A multi-ship cooperative collision avoidance decision-making model based on queuing theory framework, such as... Figure 3As shown, this model integrates ships currently in a decision-making state, neighboring ships forming an encounter situation, and an intelligent decision engine to construct a dynamic service system. Ships that have not yet created an urgent situation form a waiting queue, awaiting system scheduling and service. The core processing unit of this model includes an input process, a queuing mechanism, and a service mechanism, the specific design of which is as follows:

[0152] Input process modeling module:

[0153] Vessel queuing characteristics: In theory, it supports an unlimited number of individual vessels participating in the queue, but in practical applications, an upper limit needs to be set according to the navigation capacity of the waterway to ensure the stability and efficiency of the system.

[0154] Arrival pattern recognition: used to determine whether a ship arrives alone or in a convoy through spatiotemporal correlation analysis;

[0155] Encounter event generation pattern: The ship arrival pattern is modeled using an improved Poisson process and combined with a collision risk dual threshold triggering mechanism. When the real-time monitoring value exceeds the preset threshold, the queuing service process is automatically started. The number and status of arriving ships in the first priority are confirmed through spatiotemporal correlation analysis. Based on this, the genetic algorithm is used for encoding and optimization to ensure timely response to potential collision risks.

[0156] Queuing and Service Mechanism Module: Includes:

[0157] Queue capacity configuration submodule: Used to set the traffic carrying capacity of a limited buffer zone to simulate the actual waterway. When the number of ships exceeds the capacity threshold, external flow control measures will be triggered to prevent traffic congestion and collision accidents.

[0158] The hierarchical priority rule submodule is used to implement a dual-priority scheduling strategy. Service agencies focus on handling high-urgency vessels in the Q1 queue, adopting a preemptive service model to ensure that critical decisions can be executed first, thereby effectively reducing the risk of collision.

[0159] Service Desk Configuration and Service Module: Based on the characteristics of multiple ship encounters, a multi-layer coding genetic algorithm (such as...) is used. Figure 7 As shown, a co-evolutionary model is established, and a single service station is used to simultaneously optimize decisions for ships of the same priority. This service model can significantly improve decision-making efficiency and ensure that ships can quickly and safely leave the intersection area.

[0160] Based on the above design concept, a multi-stage collision avoidance decision-making strategy for ships was constructed, such as... Figure 4 As shown. The specific execution steps are as follows:

[0161] S31: Utilize shipboard navigation aids to monitor the ship's motion status in real time and obtain key information such as the ship's heading, speed, and position;

[0162] S32: Based on the detected ship motion state, calculate collision avoidance parameters such as relative speed, bearing, DCPA (distance to nearest encounter), and TCPA (time to nearest encounter) to provide a basis for subsequent decision-making;

[0163] S33: Based on the calculation results of the collision avoidance parameters, encountering vessels are divided into two priority gradients, Q1 and Q2. The Q1 gradient contains vessels with high urgency and requires priority handling;

[0164] S34: Check the number of ships (Num) in gradient Q1. If Num≥2, proceed to step S35; otherwise, return to step S31 and continue detecting ship motion status.

[0165] S35: Based on the number of ships (Num) in the Q1 gradient, determine the optimization object and the corresponding decision variables. These decision variables will be used in the subsequent genetic algorithm optimization process;

[0166] S36: Establish a multi-layer chromosome encoding mechanism for genetic algorithms to encode the decision variables of the ship in the Q1 gradient. This encoding method can fully express the complexity and diversity of the decision variables;

[0167] S37: Based on the objective and constraints of the collision avoidance decision, establish the objective function equation and constraint equation. Simultaneously, design suitable genetic operators to ensure that the genetic algorithm can efficiently and accurately search for the optimal solution.

[0168] S38: Genetic optimization is used to collaboratively optimize multi-objective decision schemes. Through continuous iteration and optimization, a collision avoidance decision scheme that satisfies all constraints and has the optimal objective function value is obtained.

[0169] S39: Once the vessels in gradient Q1 have completed the collision avoidance and returned to their original course, step S31 is executed again until all vessels have safely passed the clearance, at which point the entire collision avoidance decision-making process ends.

[0170] Co-evolution mechanism

[0171] The architecture of a multi-ship cooperative collision avoidance decision optimization model based on genetic evolution mechanism is as follows: Figure 5 As shown, its core innovation lies in the adoption of a hierarchical real-number encoding system. This model overcomes the limitations of traditional single-chromosome expression by constructing a multi-layered chromosome structure. Each layer of encoding corresponds to the set of decision variables for a single ship in the Q1 priority queue, achieving unified modeling of multi-ship collaborative decision-making. The specific technical implementation path is as follows:

[0172] (1) Optimization of the basic framework of genetic algorithm

[0173] SGA=(C,E,P0,M,Φ,Γ,Ψ,T e (8)

[0174] In the formula:

[0175] Encoding mapping mechanism (C): Mapping decision variables to gene sequences;

[0176] Fitness assessment (E): Quantifying the merits of individual solutions;

[0177] Population initialization (P0): Generates an initial set of solutions;

[0178] Genetic operator combinations (Φ / Γ / Ψ): perform selection, crossover, and mutation operations;

[0179] Termination determination (T) e ): Preset algorithm convergence conditions.

[0180] The standard operation process consists of six execution phases, the specific execution process is as follows: Figure 5 As shown.

[0181] ① Randomly generate a population of M individuals, each individual representing a chromosome gene encoding string;

[0182] ② Calculate the population fitness value. If the termination condition is met, output the optimal solution; otherwise, proceed with the genetic operation. Perform individual replication operation based on the fitness value. Individuals with higher fitness have a higher probability of being selected.

[0183] ③ When the roulette wheel selection operation is performed, individuals with higher fitness have a higher probability of replication;

[0184] ④ According to the preset crossover probability p c Implement multi-point crossover to generate offspring individuals;

[0185] ⑤ According to the preset mutation probability p m Perform Gaussian mutation to maintain population diversity;

[0186] ⑥ After generating a new generation of population, return to the fitness assessment phase.

[0187] (2) Implementation of multi-level coding genetic algorithm

[0188] This algorithm employs a real-number encoding scheme, extending the traditional single-layer chromosome into a multi-layer structure. Each layer of encoding corresponds to the collision avoidance decision scheme of a single ship in the first priority queue, realizing a global expression of multi-ship cooperative strategies from a single chromosome.

[0189] Individual coding strategy

[0190] The steering angle φ of each ship is encoded using floating-point numbers. i and steering operation time t i coding( Figure 6The number of chromosome layers is dynamically determined by the number of ships with the highest priority, and each layer corresponds to a single ship's decision-making scheme. Figure 7 ),make sure:

[0191] ① The original course is automatically restored after the turning operation;

[0192] ② The decision variables are directly related to the kinematic characteristics of the ship.

[0193] 1) Population initialization rules:

[0194] To meet the "early, large, wide, and clear" avoidance requirements of the International Maritime Collision Prevention Regulations, the constraints for the initialization phase are set as follows:

[0195] ① The range of turning angles for the yielding vessel: π / 12 < |φ i |<π / 3;

[0196] ② Turning time limit: 300s < t i <1800s;

[0197] 2) Fitness function design

[0198] Based on metrics such as distance, safety, path smoothness, and rerouting, a multi-objective optimization function is constructed as follows:

[0199]

[0200] In the formula:

[0201] D e Minimum meeting distance between ships;

[0202] D s : Radius of the security domain;

[0203] Ship trajectory;

[0204] τ: Path smoothness weighting coefficient;

[0205] N: Number of ships with first priority.

[0206] The genetic operators in a multi-population genetic algorithm with a multi-layered coding structure include:

[0207] ①Select Operation

[0208] The core function of the selection operator is to retain individuals with stronger survival competitiveness through a probabilistic selection mechanism, employing an improved roulette wheel selection strategy to achieve survival of the fittest in the population. The operation of this operator includes the following core steps:

[0209] S411. Determine the probability allocation mechanism;

[0210] First, calculate the probability of each individual in the population being selected in the offspring population.

[0211]

[0212] In the formula, Let N be the fitness value of the i-th individual in population k, and N be the population size. M This refers to the population size.

[0213] S412. Conduct individual screening process:

[0214] Generate a random number P1 between [0,1], and find a value that satisfies the condition. The required chromosomes, rotate N M The next roulette wheel yields a new generation of population with a size of N;

[0215] S413. Determine the elite retention strategy

[0216] To avoid losing high-quality solutions, the top 5% of individuals in terms of fitness are directly promoted to the next generation, with the remaining spots determined through a roulette wheel selection mechanism. This strategy effectively maintains the evolutionary stability of the population while preserving selection pressure.

[0217] ② Cross operations

[0218] The design of the segmented crossover operator, and the specific execution steps are as follows: Figure 8 As shown:

[0219] S421: Randomly select two chromosomes from the population and decide whether to perform the crossover operation based on the crossover probability;

[0220] S422: Randomly select the crossover point for individuals and check the feasibility of the two new individuals generated. If the requirements are not met, the crossover operation is repeated.

[0221] S423: Repeat the above steps N times to complete the crossover operation for a single population;

[0222] ③ Mutation operation

[0223] A Gaussian adaptive mutation strategy is introduced as an auxiliary method to maintain population diversity and generate new individuals, avoiding the algorithm from getting stuck in local convergence. The specific execution steps are as follows:

[0224]

[0225] S431: Randomly select the target chromosome in the current population and determine whether to perform this round of gene modification operation based on the preset mutation probability pm.

[0226] S432: After randomly determining the mutation sites, use a mean of μ and a variance of P.2 The Gaussian perturbation strategy generates new gene values, and local neighborhood search is implemented to enhance the algorithm's exploration capability, but the original gene positions generated by heuristic initialization are retained;

[0227] S433: Repeat the above operation N times to achieve population variation and update through gradual gene perturbation.

[0228] Example

[0229] Ship Encounter Scenario Settings

[0230] To verify the effectiveness of the cooperative collision avoidance decision-making model in multi-ship encounter scenarios, simulation verification experiments were conducted using two typical encounter situations. Scenario 1 simulates a three-ship interaction scenario, focusing on the basic coordinated avoidance mechanism. Specific parameters are detailed in Table 1 (including initial position coordinates, speed, and heading, etc.). Its spatial situation visualization is shown below. Figure 9 As shown in (a);

[0231] Table 1 Initial Encounter Situation Settings for Scenario 1

[0232]

[0233]

[0234] Scenario 2 is upgraded to a complex six-ship encounter situation, focusing on testing multi-objective collaborative decision-making capabilities in a high-density environment. Relevant ship kinematic parameters and initial situations are listed in Tables 2 and 3 respectively. Figure 10 (a).

[0235] Table 2 Initial Encounter Situation Settings for Scenario 2

[0236] Ships initial position / n mile Speed / kn Heading / ° S1 (0,0) 13 45 S2 (8,8) 13 225 S3 (7.7,0) 13 315 S4 (0,8) 13 135 S5 (-3.5,-1) 13 90 S6 (11.5,9) 13 270

[0237] As a test benchmark platform, the ship dimensions and maneuvering performance parameters of sample ship A are detailed in Table 3. The parameter settings take into account both the characteristics of typical merchant ships and the algorithm adaptation requirements.

[0238] Table 3. Main dimensions and maneuverability data of sample vessels

[0239]

[0240] Priority gradient partitioning

[0241] To verify the effectiveness of the gradient calculation model in ship clustering analysis, this study investigates conflict analysis for two typical encounter scenarios based on a spatiotemporal collision hazard assessment system. The specific technical verification process is as follows:

[0242] (1) Scenario 1 Verification (Three-ship Interaction Scenario)

[0243] Tables 4 and 5 show that the three vessels maintained free navigation during the initial phase. When the distance between S1 and S2 reached a dangerous threshold at t=540s, immediate action was initiated. Figure 2 The priority assessment procedure is shown below. This procedure quantifies collision risk in both space and time: when the binary comparison result of spatial collision hazard (SCR) and temporal collision hazard (TCR) is both 1, the system determines that S1 and S2 constitute a head-on conflict and automatically classifies them into the first avoidance priority echelon. After both ships perform a starboard turn to avoid the collision and resume their original course, the risk between S1 / S2 / S3 is eliminated, and no secondary cluster analysis is required.

[0244] Table 4 Scenario 1 First Priority Judgment

[0245] combination <![CDATA[r s >r0]]> <![CDATA[r t (t=t1)<r0]]> First gradient S1,S2 1 1 1 S1,S3 0 1 0 S2,S3 0 0 0

[0246] Table 5 shows the specific avoidance action plans for each vessel in Scenario 1.

[0247] Ships Action time / s Steering operation / ° New course sailing time / s S1 540 25 420 S2 540 30 360 S3 - - -

[0248] (2) Scenario 2 Verification (Six Ships Interaction Scenario)

[0249] Tables 6 and 7 show that the six ships were initially at a safe distance. At t=332s, the distance between S1 and S3 triggered a warning value, and the model initiated multi-ship collaborative analysis: through spatiotemporal risk matrix calculation, S1 / S2 / S3 / S4 were assigned to the first priority sequence. After this batch of ships completed collaborative avoidance, the system dynamically reconfigured priorities: the original multi-ship encounter problem evolved into two intersecting encounter situations, S3-S6 and S4-S5, with the corresponding ships implementing starboard maneuvers to eliminate collision risks.

[0250] Table 6 Scenario 2 First Priority Judgment

[0251]

[0252] Table 7 Specific Avoidance Action Plans for Each Vessel in Scenario 2 Table 7 Specific Avoidance Action

[0253]

[0254]

[0255] (3) Model technical characteristics

[0256] This gradient calculation system achieves hierarchical analysis of multi-ship conflict characteristics through spatiotemporal urgency quantification. Its core value lies in: 1) establishing a dynamic mapping mechanism between collision risk and avoidance priority; 2) providing a precise set of optimization objects for collaborative decision-making models; and 3) determining the physical boundary conditions for chromosome coding layers. This spatiotemporally coupled analysis paradigm effectively solves the decision-making lag problem of traditional methods in complex encounter scenarios.

[0257] Performance evaluation of decision-making schemes

[0258] To verify the effectiveness of the proposed multi-layer coded ship cooperative collision avoidance decision-making model, a simulation verification system was constructed, including a three-ship basic scenario and a six-ship complex scenario. The simulation results are as follows: Figure 9-10 As shown, the model updates the avoidance target group in real time through a dynamic priority management mechanism and uses a multi-layer real-number encoded genetic algorithm to perform global decision optimization for the first-priority vessel. Its technical implementation includes five core modules: a gradient calculation module based on spatiotemporal collision hazard, an improved roulette wheel selection operator, a segmented crossover mutation operator, a multi-objective fitness evaluation function, and a dynamic population update mechanism. Simulation results show that the optimized trajectory fully represents the entire "avoidance-return" operation process. The rudder angle-speed response curve shows that the yielding vessel's maneuvering characteristics conform to ship dynamics constraints, and the minimum encounter distance is always greater than the safe zone radius, verifying the explicit compliance of the decision scheme with the International Regulations for Preventing Collisions at Sea (COCR). Through the synergistic effect of dynamic priority management and multi-layer encoded optimization, this model achieves hierarchical analysis of multi-vehicle encounter situations and intelligent selection of decision objects. Its technical architecture can effectively support the intelligent decision-making needs in complex waterway multi-vehicle interaction scenarios.

[0259] 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 ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm, characterized in that: Includes the following steps: S1: Acquire and process relevant information of each ship in the encounter situation based on sensors; S2: Based on the collision hazard model, a dynamic early warning mechanism is constructed for multi-ship encounter scenarios: when the distance between any two ships triggers the hazard threshold, the spatial collision hazard and time urgency indicators are coupled to construct a continuous quantification priority gradient model based on the 0-1 scale. The priority gradient is generated through dynamic quantification of spatiotemporal urgency to achieve accurate clustering of collision risks and determination of collision avoidance responsibilities. S3: Based on the queuing theory framework, a multi-stage ship collision avoidance strategy is constructed. Specifically, this includes designing a parallel service architecture for multi-ship collaborative optimization, establishing a multi-ship collision avoidance decision queuing system with specific spatiotemporal coupling characteristics, introducing a dynamic time window allocation mechanism and a spatiotemporal constraint coupling network modeling method, performing multi-objective optimization on the ship decision scheme with the first priority gradient, and developing a dynamically adaptive intelligent ship collision avoidance strategy. S4: Based on the genetic evolution mechanism, a multi-ship collaborative collision avoidance decision optimization model is constructed. A multi-level coding genetic algorithm is used to integrate multi-objective adaptation functions. Through the synergistic effect of dynamic priority management and multi-level coding optimization, the decision schemes for multiple ships meeting in the first priority are determined, realizing hierarchical analysis of decision-making in multi-ship encounter situations.

2. The ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm according to claim 1, characterized in that: The collision hazard model is constructed as follows: The nearest encounter distance and the time to reach the nearest encounter distance were established using a circular ship domain model and a Gaussian fitting method. Spatial collision hazard and temporal collision hazard were established respectively, and their synthesis method was determined. The determination of the spatial collision hazard level is as follows: The risk level of a space collision between the two ships is set as R. s The membership function r is established based on the asymmetric Gaussian equation. s for: Where, take r0 = 0.5, when f s When (t) = 1, r s =0.5, which defines the spatial collision risk value when the minimum safe meeting distance between two ships is equal to the radius of the ship's territory; The determination of the time-based collision risk level is as follows: The time-collision risk set between this vessel and other vessels is R. t R is established through the asymmetric Gaussian equation. t Membership function r t for: Where, when f t When (t) = 1, define r t =0.5 is the critical value of the collision risk level at the time when the encounter situation is formed, that is, the time membership function value corresponding to the two ships approaching from the applicable distance to the safe encounter distance in a specific encounter situation; The synthesis method is determined by synthesis rules, which represent the actual collision risk between ships by synthesizing spatial collision risk and temporal collision risk. It is a composition operator: ① When r s < r < 0, it is considered that there is no risk of collision between ships and no action needs to be taken; ② When r s ≥ r0, r t < r0, it is considered that there is no risk of collision between ships and no action needs to be taken; ③ When r s ≥r0, r t When r ≥ 0, there is a risk of collision between vessels, and the vessel giving way needs to take evasive action.

3. The ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm according to claim 1, characterized in that: The coupling of spatial collision hazard and time urgency indices constructs a continuous quantized priority gradient model based on a 0-1 scale. The priority gradient is generated through dynamic quantization of spatiotemporal urgency, enabling precise clustering of collision risks and delineation of collision avoidance responsibility. The process is as follows: S21: Use the shipborne navigation system to obtain dynamic and static data of each vessel; S22: Initialize the primary priority queue Q1 and the secondary priority queue Q2 to be empty sets; S23: Initialize the index parameter k = 1, and construct the pairing pointer q = k + 1; S24: Calculate the spatiotemporal composite hazard index (SCR) for the current vessel i=k and the candidate vessel j=q. ij Time Collision Risk (TCR) ij ; S25: When SCR is satisfied ij ∈[r0+Δr,1] and TCR ij If both ∈[r0+Δr,1] are true, then the ship pair [i,j] is added to queue Q1; otherwise, proceed to step S26. S26: Incrementing paired pointer q = q + 1; S27 If q≤N, continue to step (4); otherwise check the state of index k; if k≥N, construct Q2=P-Q1, otherwise increment k=k+1 and reset q=k+1 and return to step (3); (8) When the full ship traversal is completed and k≥N, a secondary priority queue Q2=P-Q1 is generated through international verification operation; (9) Finally, output the classification results of the dual priority queues Q1 (high priority) and Q2 (second priority).

4. The ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm according to claim 1, characterized in that: The establishment of a multi-ship collision avoidance decision queuing system with specific spatiotemporal coupling characteristics includes: Input process modeling module: used to determine whether a ship arrives alone or in a convoy through spatiotemporal correlation analysis; An improved Poisson process is used to model the ship arrival pattern, combined with a dual-threshold triggering mechanism for collision risk. When the real-time monitoring value exceeds the preset threshold, the queuing service process is automatically initiated to confirm the number and status of arriving ships in the first priority, ensuring timely response to potential collision risks. Queuing and Service Mechanism Module: Includes Queue capacity configuration submodule: Used to set the traffic carrying capacity of a limited buffer zone to simulate the actual waterway. When the number of ships exceeds the capacity threshold, external flow control measures will be triggered to prevent traffic congestion and collision accidents. The hierarchical priority rule submodule is used to implement a dual-priority scheduling strategy. Service agencies focus on handling high-urgency vessels in the Q1 queue, adopting a preemptive service model to ensure that critical decisions can be executed first, thereby effectively reducing the risk of collision. Service Desk Configuration and Service Module: Based on the characteristics of multiple ship encounters, a collaborative optimization model is established using a multi-layer coded genetic algorithm, and a single service desk is used to perform simultaneous decision optimization for ships with the same priority.

5. The ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm according to claim 1, characterized in that: The multi-layer coding genetic algorithm employs a real-number coding scheme, expanding a single-layer chromosome into a multi-layer structure. Each layer of coding corresponds to the collision avoidance decision scheme of a single ship in the first priority queue, realizing the global expression of multi-ship cooperative strategies from a single chromosome. The individual coding strategy is as follows: Floating-point encoding is used to encode the turning angle φi and turning operation time ti for each ship; the number of chromosome layers is dynamically determined by the number of ships with the highest priority, and each layer corresponds to a single ship's decision-making scheme, ensuring that: The original course is automatically restored after a turning maneuver. Decision variables are directly related to the kinematic properties of the ship; The population initialization rule has the following constraints during the initialization phase: The turning angle range for yielding vessels is: π / 12 < |φ i |<π / 3; Turning time limit: 300s < t i <1800s.

6. The ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm according to claim 1, characterized in that: The multi-objective adaptation function is constructed based on distance, safety, path smoothness, and rerouting indicators, as detailed below: In the formula: D e Minimum meeting distance between ships; D s : Radius of the security domain; Ship trajectory; τ: Path smoothness weighting coefficient; N: Number of ships with first priority.

7. The ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm according to claim 1, characterized in that: The genetic operation operators of the multi-layer coded genetic algorithm include selection operation operators; The operation flow of the selection operator includes the following core steps: S411: Determine the probability allocation mechanism; First, calculate the probability of each individual in the population being selected in the offspring population. In the formula, F i k Let N be the fitness value of the i-th individual in population k, and N be the population size. M Population size; S412: Conduct individual screening; Generate a random number P1 between [0,1], and find a number that satisfies P1. <F i k The required chromosomes, rotate N M The next roulette wheel yields a new generation of population with a size of N; S413: Determine the elite retention strategy; The top 5% of individuals in terms of fitness are directly admitted to the next generation, and the remaining spots are determined through a roulette-like competition mechanism.

8. The ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm according to claim 1, characterized in that: The genetic operation operators of the multi-layer coding genetic algorithm include the crossover operation operator, and the specific execution steps of the crossover operator are as follows: S421: Randomly select two chromosomes from the population and decide whether to perform the crossover operation based on the crossover probability; S422: Randomly select the crossover point for individuals and check the feasibility of the two new individuals generated. If the requirements are not met, the crossover operation is repeated. S423: Repeat the above steps N times to complete the crossover operation for a single population.

9. The ship cooperative collision avoidance decision-making method based on a multi-layer coded genetic algorithm according to claim 1, characterized in that: The genetic operation operator of the multi-layer coding genetic algorithm includes a mutation operation, and the execution steps of the mutation operation are as follows: S431: Randomly select the target chromosome in the current population and determine whether to perform this round of gene modification operation based on the preset mutation probability pm. S432: After randomly determining the mutation sites, use a mean of μ and a variance of P. 2 The Gaussian perturbation strategy generates new gene values, and local neighborhood search is implemented to enhance the algorithm's exploration capability, while retaining the original gene positions generated by heuristic initialization; S433: Repeat the above operation N times to achieve population variation and renewal through gradual gene perturbation.

Images