Direct ship collision avoidance decision-making method based on way-giving ship intention reasoning

Abstract

The invention relates to a direct navigation ship collision avoidance decision-making method based on way-giving ship intention reasoning. Comprising the steps of dividing collision risk grades according to collision risk indexes of a direct navigation ship and a way-giving ship for relative movement of the direct navigation ship and the way-giving ship; constructing a cooperation relation matrix under a global meeting situation within the observable range of the direct navigation ship, judging whether the space-time effect difference between the predicted state and the actual observation state of each way-giving ship under the visual angle of the direct navigation ship exceeds a set threshold value or not, if yes, judging that the way-giving ship intends to be not cooperative, adjusting the cooperation relation matrix, and if not, judging that the way-giving ship does not cooperate with the set threshold value; judging whether the collision risk index exceeds a set threshold value corresponding to the collision risk level or not, and if yes, triggering the direct navigation ship to execute an avoidance action; and optimizing the safety objective function and the economy objective function through an NSGA-II algorithm to obtain an optimal scheme. According to the invention, the intelligence, precision and practicability of collision avoidance decision making are obviously improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of ship intelligent collision avoidance decision-making, and particularly relates to a collision avoidance decision-making method for a stand-on ship based on the inference of the give-way ship's intention. Background Art

[0002] With the continuous growth of maritime traffic flow, ship collision avoidance faces increasingly complex challenges. In a dynamic and complex navigation environment, insufficient situation awareness, limited decision-making space, and the complexity of multi-ship encounters seriously affect the effectiveness of collision avoidance operations. The International Regulations for Preventing Collisions at Sea provide basic principles, but in actual navigation, it is not realistic to assume that all ships strictly comply with the rules. In particular, Rule 17 stipulates that a stand-on ship shall maintain its course and speed in the initial stage of an encounter, but when the give-way ship fails to take effective avoidance actions, the stand-on ship shall react in a timely manner.

[0003] Currently, the collision avoidance decision-making of a stand-on ship mainly relies on the subjective judgment of the crew, which is prone to increasing risks due to misjudgment. Accidents show that relying solely on rule constraints is not sufficient to ensure navigation safety.

[0004] To prevent similar accidents, a stand-on ship needs to have the ability to accurately infer the intention of the give-way ship and be able to quickly take coordinated and effective collision avoidance measures. This is not only the core requirement for ensuring navigation safety but also directly related to the economy and efficiency of navigation. Therefore, developing technologies to enhance the situation awareness and collision avoidance decision-making ability of stand-on ships has important theoretical and practical significance.

[0005] Existing research mainly focuses on ship intention estimation and collision avoidance decision-making optimization, but there are still limitations in practical applications. In terms of intention estimation, the index-based method evaluates the collision risk through DCPA, TCPA, inter-ship distance, and relative speed. Although simple and intuitive, it lacks a unified weight standard and is difficult to handle complex dynamic situations; the causality-based method (such as the Bayesian model) relies on historical AIS data or predefined rules and is applicable to static environments, but its effect is limited in dynamically changing situations. Existing research should focus more on high-risk situations and improve adaptability and accuracy by combining real-time navigation data.

[0006] In terms of collision avoidance decision-making optimization, existing methods such as optimization algorithms, virtual vector methods, and artificial intelligence technologies, although having advantages in dynamic collision avoidance, mostly focus on the risk elimination strategy of the give-way ship and pay less attention to the optimization of the stand-on ship's actions. In addition, these methods do not fully consider the impact of ship maneuverability on decision-making, resulting in the stand-on ship's decision being difficult to effectively avoid collision risks.

[0007] Therefore, in a complex multi-ship encounter situation, it is necessary to optimize the collision avoidance decision-making of a stand-on ship by combining accurate intention estimation, and fully consider the changes in ship handling characteristics and dynamic environment to ensure navigation safety and efficiency. Summary of the Invention

[0008] The object of the present invention is to solve the deficiencies existing in the above-mentioned background technology, and provide a collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention. The focus is on optimizing the action timing and strategy of the stand-on vessel in a complex navigation environment. By inferring the intention of the give-way vessel and quantifying the collision stage, the optimal action timing of the stand-on vessel is accurately determined, and an efficient collision avoidance decision is formulated in combination with its maneuvering ability, so as to improve the economy while ensuring navigation safety. This technology is particularly applicable to the development of intelligent navigation systems in busy sea traffic areas and dynamic complex environments, providing effective technical support for improving the scientificity and efficiency of ship collision avoidance decision-making.

[0009] The technical solution adopted by the present invention is: a collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention, including the following steps:

[0010] For the relative motion between the stand-on vessel and the give-way vessel, the collision risk level is divided according to the collision risk index (CRI) between the stand-on vessel and the give-way vessel;

[0011] A cooperation relationship matrix under the global encounter situation is constructed within the observable range of the stand-on vessel, and it is judged whether the spatio-temporal effect difference between the predicted state and the actual observed state of each give-way vessel from the perspective of the stand-on vessel exceeds a set threshold. If so, the intention of the give-way vessel is determined to be non-cooperative, the cooperation relationship matrix is adjusted, and it is judged whether the collision risk index exceeds the set threshold corresponding to the collision risk level. If so, the stand-on vessel is triggered to execute an avoidance action;

[0012] After the avoidance action is triggered, the nonlinear velocity obstacle (NL-VO) method is used to calculate the collision avoidance decision risk, and a safety objective function is constructed. An economic objective function is constructed based on the ship speed and turning radius. The safety objective function and the economic objective function are optimized through the NSGA-II algorithm to obtain the optimal collision avoidance decision.

[0013] In the above solution, the "global encounter situation" refers to the overall maritime traffic situation formed by all target vessels (give-way vessels) within the observable range of the stand-on vessel from the perspective of the stand-on vessel.

[0014] The collision risk index (CRI) is obtained by normalizing the degree of domain intrusion (DDV) and the remaining time of domain intrusion (TDV). The collision risk level between the stand-on vessel and the give-way vessel is divided into four stages: cautious, warning, dangerous, and extremely dangerous according to the value of the collision risk index (CRI). The collision risk index CRI is determined by the following formula:

[0015] ;

[0016] where m is the collision risk adjustment factor for the degree of domain intrusion; n is the collision risk adjustment factor for the remaining time of domain intrusion; , ; DDV is the degree of domain intrusion; TDV is the remaining time of domain intrusion, and the value range is between , and the closer the value is to 1, the higher the collision risk.

[0017] The degree of domain intrusion (DDV) and the remaining time of domain intrusion (TDV) are calculated respectively using the eccentric ellipse domain model. The method is as follows: Set the length of the major semi-axis, the length of the minor semi-axis, and the corresponding offset of the eccentric ellipse domain according to the hull size and speed of the through ship; Substitute the relative position of the give-way ship with respect to the through ship into the eccentric ellipse domain equation. If the intrusion judgment formula is satisfied, it indicates that the give-way ship has intruded into the domain; Solve the relative motion equation of the give-way ship and the through ship to obtain the time root for the give-way ship to enter or leave the eccentric ellipse domain, and thereby determine the degree of domain intrusion (DDV) and the remaining time of domain intrusion (TDV), which are used to characterize the depth and urgency of the collision risk.

[0018] The spatio-temporal effect difference between the predicted state and the actual observed state of the give-way ship from the perspective of the through ship is obtained through the following steps: Construct the set of actual observed states of the through ship and the give-way ship. Based on the decision-making and observation information after a period of time, generate the set of predicted states of the give-way ship at a certain moment after a period of time, and obtain the predicted space effect coefficient and the predicted time effect coefficient through data processing; After a period of time, obtain the new set of actual observed states of the give-way ship, and obtain the actual observed space effect coefficient and the actual observed time effect coefficient through data processing; Use the BPA function to obtain the spatio-temporal effect corresponding to the predicted state and the spatio-temporal effect corresponding to the actual observed state from the predicted space effect coefficient, the predicted time effect coefficient, the actual observed space effect coefficient, and the actual observed time effect coefficient, and fuse the spatio-temporal effect corresponding to the predicted state and the spatio-temporal effect corresponding to the actual observed state.

[0019] The spatio-temporal effect difference between the predicted state and the actual observed state of the give-way ship from the perspective of the through ship after a period of time is expressed as:

[0020] ;

[0021] ;

[0022] ;

[0023] Among them, represents the confidence level that the predicted state of the give-way ship from the perspective of the through ship at the current moment supports cooperation after a period of time ; represents the confidence level that the predicted state of the give-way ship from the perspective of the through ship at the current moment does not support cooperation after a period of time ; Indicates the current moment A period of time later The degree to which the collision avoidance behavior given by the predicted state of the give-way vessel from the perspective of the stand-on vessel is recognized as supported or not supported; Indicates the current moment A period of time later The confidence level that the actual observed state of the give-way vessel from the perspective of the stand-on vessel supports cooperation; Indicates the current moment A period of time later The confidence level that the actual observed state of the give-way vessel from the perspective of the stand-on vessel does not support cooperation; Indicates the current moment A period of time later The degree to which the collision avoidance behavior given by the actual observed state of the give-way vessel from the perspective of the stand-on vessel is recognized as supported or not supported; Indicates the current moment A period of time later The spatio-temporal effect corresponding to the actual observed state of the give-way vessel from the perspective of the stand-on vessel; Indicates the current moment A period of time later The spatio-temporal effect corresponding to the predicted state of the give-way vessel from the perspective of the stand-on vessel; Indicates the current moment A period of time later The spatio-temporal effect difference between the predicted state and the actual observed state of the give-way vessel from the perspective of the stand-on vessel.

[0024] Cooperation relationship matrix Determined by the following formula:

[0025] ;

[0026] Wherein, Are respectively the cooperation relationships between the 1st, 2nd, 3rd... nth give-way vessels and the stand-on vessel from the perspective of the stand-on vessel. The value 1 represents cooperation, and the value 0 represents non-cooperation.

[0027] Constructing the safety objective function includes the following steps: mapping the relative positions, speeds, and headings of the stand-on vessel and the give-way vessels to the velocity space, quantifying the spatial danger by judging the distance between the velocity vector of the stand-on vessel and the boundary of the non-linear velocity obstacle area, and quantifying the time urgency by the distance to the vertex of the collision triangle; calculating the deviation between the current heading of the stand-on vessel and the optimal velocity heading, and weighted quantifying the heading deviation risk by combining the included angle of the left and right maximum variable velocity directions; integrating and quantifying the spatial danger, time urgency, and heading deviation risk to obtain the safety objective function.

[0028] The safety objective function is determined by the following formula:

[0029] Safety objective function f1:

[0030] ;

[0031] wherein, represents the distance between the current speed and the boundary of the non-linear speed obstacle zone; represents the distance between the current speed and the starting point of the collision triangle ; represents the risk quantification of the course; are respectively , and weighting coefficients of.

[0032] The economic objective function is determined by the following formula:

[0033] Economic objective function f2: ;

[0034] wherein, represents the ship's speed; represents the turning radius in the collision avoidance operation; decision variables , , respectively represent the speed change ratio, the course change amount, and the avoidance duration; rad() represents radians.

[0035] After triggering the avoidance action, it also includes: defining the collision avoidance decision as three main decision variables: speed change ratio, course change amount, and avoidance duration, and using the Nomoto model to evaluate the response of the oncoming ship to these decision variables.

[0036] The beneficial effects of the present invention are as follows:

[0037] 1. By introducing the degree of domain invasion (DDV) and the remaining time of domain invasion (TDV) into the collision risk calculation, and using the Dempster-Shafer (D-S) evidence theory to judge whether the give-way vessel is cooperative, the system can automatically and quickly identify risks and infer the intentions of the give-way vessel in the encounter situation, improving the automation and intelligence level of navigation collision avoidance.

[0038] 2. By constructing a global collision risk degree matrix and a cooperation relationship matrix, and triggering the oncoming ship to take the initiative to avoid when detecting that some give-way vessels are "uncooperative" and there is a high collision risk, corresponding risk assessments and intention judgments can be made for different give-way vessels in the complex situation of multi-ship encounters, reflecting the self-adaptability of the method and its flexible response ability in a multi-objective environment.

[0039] 3. By adopting the NSGA-II algorithm, the safety objective function and the economic objective function are simultaneously incorporated into the optimization process, enabling the stand-on vessel to not only minimize the collision risk to the greatest extent during multi-ship collision avoidance but also minimize the deviation from the course or speed loss as much as possible, ensuring the overall operation efficiency of the vessel and presenting multiple optional solutions under the Pareto front solution, providing more flexible and practical decision-making support for real navigation scenarios.

[0040] 4. The Nomoto model is used to evaluate the maneuvering response ability of the stand-on vessel during the decision-making process, which can consider the actual operation constraints, turning ability, and speed dynamic characteristics to ensure the physical feasibility of the collision avoidance strategy; the Nonlinear Velocity Obstacle (NL-VO) method identifies the collision risk for various possible motion modes (such as non-constant speed straight navigation) that the give-way vessel may adopt, simplifies the dependence on high-precision prediction of the other party's trajectory, and improves the applicability and robustness in complex environments.

[0041] 5. The multi-objective evolutionary optimization (NSGA-II) has high computational efficiency; coupled with the relatively controllable computational workload of the Nomoto model and NL-VO itself, it is conducive to collision risk monitoring and decision update in real-time or near-real-time environments and is suitable for integrated application in modern intelligent ship bridge systems or autonomous navigation systems.

[0042] 6. The D-S evidence theory is combined to judge in real-time whether the give-way vessel complies with the rule; the stand-on vessel can only maneuver actively at necessary moments, which not only conforms to the regulations but also takes into account the actual operation, reducing the potential liability and risk arising outside the international rules.

[0043] 7. During multi-ship rendezvous, through the real-time assessment of the collision risk, the dynamic identification of the give-way vessel's intention, and the dual optimization of safety and economy, the intelligence, accuracy, and practicality of the collision avoidance decision are significantly improved, and it has good application prospects and promotion value in the scenario of multi-ship cooperative collision avoidance at sea. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is the overall flowchart of a collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention of the present invention;

[0045] Figure 2 is a schematic diagram for calculating the loss of the optimized route for collision avoidance decision-making;

[0046] Figure 3 is a schematic diagram for calculating the collision risk by the Nonlinear Velocity Obstacle method;

[0047] Figure 4 is the track chart of multiple ships in Scenario 2.1;

[0048] Figure 5 is the navigation state chart of multiple ships in Scenario 2.1;

[0049] Figure 6 It is the collision risk diagram in Scenario 2.1;

[0050] Figure 7 It is the non - linear velocity obstacle analysis diagram in Scenario 2.1;

[0051] Figure 8 It is the track diagram of multiple ships in Scenario 2.2;

[0052] Figure 9 It is the navigation state diagram of multiple ships in Scenario 2.2;

[0053] Figure 10 It is the collision risk diagram in Scenario 2.2;

[0054] Figure 11 It is the non - linear velocity obstacle analysis diagram in Scenario 2.2;

[0055] Figure 12 It is the track diagram of multiple ships in Scenario 2.3;

[0056] Figure 13 It is the navigation state diagram of multiple ships in Scenario 2.3;

[0057] Figure 14 It is the collision risk diagram in Scenario 2.3;

[0058] Figure 15 It is the non - linear velocity obstacle analysis diagram in Scenario 2.3;

[0059] Figure 16 It is the collision risk and voyage loss diagram before, during and after the avoidance opportunity in each experimental scenario. Detailed implementation manners

[0060] The following further describes the detailed implementation manners of the present invention with reference to the accompanying drawings. It should be noted here that the description of these implementation manners is used to help understand the present invention, but does not constitute a limitation on the present invention. In addition, the technical features involved in the various implementation manners of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0061] As Figure 1 shown, the present invention provides a collision avoidance decision - making method for a stand - on vessel based on the inference of the give - way vessel's intention, including the following steps:

[0062] For the relative motion between the stand - on vessel and the give - way vessel, divide the collision risk level according to the collision risk index (CRI) of the stand - on vessel and the give - way vessel;

[0063] Construct a cooperation relationship matrix in the global encounter situation within the observable range of the stand-on vessel, and determine whether the spatio-temporal effect difference between the predicted state and the actual observed state of each give-way vessel from the perspective of the stand-on vessel exceeds the set threshold. If so, determine that the intention of the give-way vessel is non-cooperative, adjust the cooperation relationship matrix, and determine whether the collision risk index exceeds the set threshold corresponding to the collision risk level. If so, trigger the stand-on vessel to execute an avoidance action;

[0064] After triggering the avoidance action, define the collision avoidance decision as three main decision variables: the speed change ratio, the course change amount, and the avoidance duration. Use the Nomoto model to evaluate the response of the stand-on vessel to these decision variables, calculate the collision avoidance decision risk using the non-linear velocity obstacle (NL-VO) method, and construct a safety objective function. Based on the ship speed and turning radius, construct an economic objective function, and optimize the safety objective function and the economic objective function through the NSGA-II algorithm to obtain the optimal collision avoidance decision.

[0065] In the above technical solution, the collision risk index (CRI) is obtained by normalizing the degree of domain intrusion (DDV) and the remaining time of domain intrusion (TDV).

[0066] In the above technical solution, use the eccentric ellipse domain model to calculate the degree of domain intrusion (DDV) and the remaining time of domain intrusion (TDV) respectively. The method is as follows: Set the length of the major semi-axis, the length of the minor semi-axis, and the corresponding offset of the eccentric ellipse domain according to the hull size and speed of the stand-on vessel; Substitute the relative position of the give-way vessel relative to the stand-on vessel into the eccentric ellipse domain equation. If the intrusion judgment formula is satisfied, it indicates that the give-way vessel intrudes into this domain; Solve the relative motion equation of the give-way vessel and the stand-on vessel to obtain the time root when the give-way vessel enters or leaves the eccentric ellipse domain, and thereby determine the degree of domain intrusion (DDV) and the remaining time of domain intrusion (TDV), which are used to characterize the depth and urgency of the collision risk.

[0067] In this embodiment, the dynamic collision risk is calculated using the degree of domain intrusion (DDV) and the remaining time of domain intrusion (TDV). The collision risk is divided through the collision risk index (CRI), providing an accurate timing reference for the effective collision avoidance measures of the stand-on vessel.

[0068] In this embodiment, the degree of domain intrusion (DDV) and the remaining time of domain intrusion (TDV) are calculated based on the eccentric ellipse domain. The parameters of this ellipse domain include the length of the major semi-axis , the length of the minor semi-axis , the offset in the direction of the major semi-axis , and the offset in the direction of the minor semi-axis , and these parameters are related to the length of the stand-on vessel . , , , .

[0069] Assume that the initial positions of the give-way vessel and the stand-on vessel are respectively and , and their headings are respectively and , and their speeds are respectively and . A local coordinate system is established with the current coordinates of the stand-on vessel as the origin As Figure 2 shown, the updated position of the give-way vessel in this coordinate system is as shown in Equation (1):

[0070] (1)

[0071] where the abscissa change of the give-way vessel , the ordinate change of the give-way vessel , and the change in the heading of the give-way vessel in the local coordinate system is .

[0072] In the domain model, if Equation (2) is satisfied, the give-way vessel begins to intrude into the domain of the stand-on vessel:

[0073] (2)

[0074] Solving for the extreme point of Equation (2) represents that the scaling factor size tangent to the domain satisfies:

[0075] (3)

[0076] If when , it means that the give-way vessel will not intrude into the domain of the stand-on vessel; if , then there is:

[0077] (4)

[0078] According to the definition of , which is the remaining time before the give-way vessel intrudes into the domain of the stand-on vessel, it is calculated by Equation (1). According to the two solutions and of the quadratic equation of one variable, the following situations can be determined:

[0079] 1) If and : The domain has been intruded and left;

[0080] 2) If And : The area is currently being invaded;

[0081] 3) If And : The area will be invaded in the future.

[0082] Indicates , Indicates the remaining time to leave the direct - course ship area, Characterizes the severity of the invasion of the give - way ship, Then reflects the time urgency. Under reasonable circumstances, The larger, The smaller, the higher the collision risk.

[0083] Using the Collision Risk Index (CRI) combines the degree of area invasion ( ) and the remaining time of area invasion ( ). Referring to the electric - field theory, Is similar to "the electric - charge quantity of the field source", while Indicates "the distance between the field source and the observation point". The definition of is as follows in Equation (5):

[0084] (5)

[0085] Where, m is the collision - risk adjustment factor for the degree of area invasion; n is the collision - risk adjustment factor for the remaining time of area invasion; , . 、 、 The value ranges of are between . The closer the value is to 1, the higher the collision risk. When , , indicating that there is no immediate collision risk.

[0086] Based on the value of the Collision Risk Index , the collision risk is divided into four stages: cautious, warning, dangerous, very dangerous, as shown in Table 1. The detailed description is as follows:

[0087] Table 1 Collision - risk level classification

[0088] In the above technical scheme, the difference in spatiotemporal effect between the predicted state and the actual observed state of the give-way ship from the perspective of the direct-navigation ship is obtained through the following steps: constructing a set of actual observed states of the direct-navigation ship and the give-way ship, generating a set of predicted states of the give-way ship at a certain moment after a period of time based on the decision and observation information after a period of time, and obtaining a predicted spatial effect coefficient and a predicted time effect coefficient through data processing; obtaining a new set of actual observed states of the give-way ship after a period of time, and obtaining an actual observed spatial effect coefficient and an actual observed time effect coefficient through data processing; obtaining the predicted spatial effect coefficient and the predicted time effect coefficient, the actual observed spatial effect coefficient and the actual observed time effect coefficient through the BPA function to obtain the spatiotemporal effect corresponding to the predicted state and the spatiotemporal effect corresponding to the actual observed state, and fusing the spatiotemporal effect corresponding to the predicted state with the spatiotemporal effect corresponding to the actual observed state.

[0089] In this embodiment, the reasoning of the cooperation relationship matrix between the OS and the TS is based on repeated verification and adjustment of the decision-making and operation behavior of the TS. This process is achieved by comparing the relative state sets of the TS and the OS. Specifically, this embodiment defines two state sets: the actual observed state set and the actual observed state set. and the predicted state set .

[0090] In time At time , the actual observed state set of the direct ship (OS) and the give-way ship (TS) is expressed as ,in: , Due to time The prediction accuracy of the give-way ship status is crucial for the subsequent cooperative relationship matrix reasoning and decision making, so another set of status sets needs to be constructed for more in-depth analysis.

[0091] To this end, this embodiment introduces a prediction state set, which is expressed as ,in, yes The status of the direct ship after time is determined by time The decision of direct sailing ship is obtained; From time Prediction from the perspective of a direct ship The state of the avoidance ship after time. , the set of predicted states is equal to the set of actual observed states.

[0092] The initial cooperative relationship matrix between the direct-travel ship and any other ship is determined by the International Regulations for Preventing Collisions at Sea, and is initially set as cooperative ( ), after the time At a certain moment When, the predicted value of the cooperation relationship matrix is obtained from the of the give-way vessel ( and . This predicted value reflects the decision-making trend of the give-way vessel. According to Dempster-Shafer (D-S) evidence theory, a finite non-empty set is assumed as the frame of discernment (FoD) of the evidence theory, which consists of two mutually exclusive hypotheses. It is defined as: , where indicates that the behavior of the give-way vessel supports the current cooperation relationship, while is the opposite, indicating non-support. The power set of is , where is the universal set and is the empty set. Let be the basic probability assignment (BPA) function defined on and respectively quantify the confidence levels that the behavior is classified as supporting or not supporting the current cooperation relationship. In particular, specifies the degree to which a given collision avoidance behavior is identified as supporting or not supporting. According to the vessel states and of the two vessels at , the distance between the two vessels can be expressed as . Based on the calculation of DDV and TDV, the degree of domain intrusion of the two vessels can be expressed as , and the remaining time of domain intrusion can be expressed as and are set to larger values if the state data of the two vessels shows that the two vessels are moving away from each other and there is no tendency to approach. The spatial proximity index between the stand-on vessel and the give-way vessel is and weighted average, expressed as shown in Equation (6):

[0093] (6)

[0094] For the give-way vessel, it is also necessary to calculate the predicted spatial efficiency factor ( ) and the predicted temporal efficiency factor ( ) for its predicted state, and the predicted After a certain time, it can be expressed as:

[0095] (7)

[0096] (8)

[0097] These two values can represent the degree of similarity in the spatial and temporal effects between the predicted state of the give-way vessel from the perspective of the stand-on vessel after a certain time and the current state. After a certain time, it can be expressed as:

[0098] And corresponding to the predicted state is The actual observed state after a certain time, expressed as , the Observed Spatial Efficiency Factor ( ), and the Observed Temporal Efficiency Factor ( ). The actual observation After a certain time, it is expressed as:

[0099] (9)

[0100] (10)

[0101] Based on the differences between these two sets of evidence, the intention of the give-way vessel is inferred, and then the cooperation relationship matrix is adjusted. Therefore, a quantitative method is needed to describe this difference:

[0102] The predicted value of the give-way vessel obtained from the cooperation relationship matrix is expressed as

[0103] (11)

[0104] The actual observed value is expressed as:

[0105] (12)

[0106] Therefore, in The spatio-temporal effect difference between the predicted state and the actual observed state of the give-way vessel from the perspective of the stand-on vessel after a certain time can be expressed as:

[0107] (13)

[0108] Among them, Represents the confidence level that the predicted state of the give-way vessel from the perspective of the stand-on vessel supports cooperation after a certain period of time after the current moment ; Represents the current moment The latter period of time The confidence level that the predicted state of the give-way vessel does not support cooperation from the perspective of the stand-on vessel; Indicates the current moment The latter period of time The degree to which the collision avoidance behavior given by the predicted state of the give-way vessel is recognized as supportive or not supportive from the perspective of the stand-on vessel; Indicates the current moment The latter period of time The confidence level that the actual observed state of the give-way vessel supports cooperation from the perspective of the stand-on vessel; Indicates the current moment The latter period of time The confidence level that the actual observed state of the give-way vessel does not support cooperation from the perspective of the stand-on vessel; Indicates the current moment The latter period of time The degree to which the collision avoidance behavior given by the actual observed state of the give-way vessel is recognized as supportive or not supportive from the perspective of the stand-on vessel; Indicates the current moment The latter period of time The spatio-temporal effect corresponding to the actual observed state of the give-way vessel from the perspective of the stand-on vessel; Indicates the current moment The latter period of time The spatio-temporal effect corresponding to the predicted state of the give-way vessel from the perspective of the stand-on vessel; Indicates the current moment The latter period of time The spatio-temporal effect difference between the predicted state and the actual observed state of the give-way vessel from the perspective of the stand-on vessel.

[0109] In this embodiment, the action trigger mechanism of the stand-on vessel includes two links. The first is to extend the collision risk calculation method to the global to form a collision risk matrix from the perspective of the stand-on vessel The definition is as shown in the formula. The matrix is a matrix of is the number of within 6 nautical miles of the distance from .

[0110] (14)

[0111] Among them, are respectively the collision risks between the 1st, 2nd, 3rd... nth give-way vessels and the stand-on vessel from the perspective of the stand-on vessel.

[0112] The second is to draw on the construction idea of and construct a cooperation relationship matrix according to the current global encounter situation , and the definition is as shown in the formula. The matrix is also The matrix, where the value 1 represents cooperation and the value 0 represents non - cooperation.

[0113] (15)

[0114] Among them, are respectively the cooperation relationships between the 1st, 2nd, 3rd... nth give - way vessels and the stand - on vessel from the perspective of the stand - on vessel. The value 1 represents cooperation and the value 0 represents non - cooperation.

[0115] and are related. As the navigation parameters of the give - way vessels within the detection range of the stand - on vessel change continuously, the stand - on vessel infers the intentions of the give - way vessels by analyzing the observable data. When the difference obtained from (13) is greater than the set evidence threshold (the threshold in this embodiment is 0.77), the variable in formula (15) changes from 1 to 0. And when the collision risk index reaches the set collision avoidance threshold corresponding to the collision risk level (0.6 in this embodiment), which is the third stage of collision avoidance, "danger", and both conditions are met, the stand - on vessel can start to execute the avoidance action.

[0116] When the stand - on vessel starts to act, it optimizes from the alternative solutions to achieve multi - ship collision avoidance. To conform to the actual navigation scenario of marine vessels and achieve multi - ship collision avoidance, based on the NSGA - II algorithm, two objective functions of safety and economy are designed for optimization, and optimization is carried out during the initialization of the algorithm population and the iterative evolution process to improve the global search ability and convergence speed of the algorithm.

[0117] In this embodiment, the ship collision avoidance decision - making is modeled as a multi - objective optimization problem, with safety and economy as the core objectives, and the NSGA - II algorithm is used for optimization. During the ship's navigation process, the avoidance action will be described by three main decision variables: (1) the speed change ratio (that is, the original speed multiplied by a certain ratio), (2) the course change , (3) the avoidance duration ,

[0118] As Figure 2 shown, represents that the stand - on vessel reaches the decision - making cycle. The initial position of the stand - on vessel is at point . At time (point B), it reaches the action time point, and the stand - on vessel starts to act according to the decision variables. The ship adjusts its speed according to the speed change ratio , adjusts its course to avoid the give - way vessel according to the course change , sails for duration, and reaches At a certain point, after passing and clearing, it resumes its original course to reach Point D, resumes its original navigation state, and continues to sail until the avoidance duration ends and the next decision-making cycle is reached. 。

[0119] Ship collision avoidance actions mainly consider safety and economy. Generally, the safer the collision avoidance decision, the worse the economy. These two goals are contrary to each other, constituting the conditions of Pareto.

[0120] To evaluate the effectiveness of the current strategy from the perspective of the actions of the stand-on vessel, this embodiment adopts the Nomoto model. On the one hand, the stand-on vessel needs to consider all possible operations that the give-way vessel may take, such as changing course, adjusting speed, or a combination of both, to avoid conflicts. On the other hand, the maneuvering ability of the stand-on vessel also directly affects its decision-making process. Given the concise linear form and fast calculation advantages of the Nomoto model, it is widely used in the simulation of ship turning ability and the prediction of achievable speed, and is particularly suitable for scenarios with low accuracy requirements but fast response. While ensuring computational efficiency, this model can provide effective ship trajectory prediction.

[0121] (16)

[0122] Among them, is the roll angular acceleration, is the roll angular velocity ( ), is the rudder angle during steering, 。 and are respectively called the time constant and the gain constant.

[0123] Assume that the current operating rudder angle is , after maintaining for a time, the course change of the ship is:

[0124] (17)

[0125] When the time is small enough, the position of the ship will be restricted within a limited area, and this area is called the "reachable domain".

[0126] The Nonlinear Velocity Obstacle (NL-VO) algorithm is used to calculate the risk of collision avoidance decisions. The NL-VO method projects the relative position, speed, and course information of the encountering ships into the velocity domain to determine whether the current speed enters the conflict area, and then evaluates the risk of collision avoidance decisions. The Nonlinear Velocity Obstacle method can handle the situation where the give-way vessel sails in a straight line at a non-constant speed, and only requires the track information of the give-way vessel, effectively improving the adaptability of strategy evaluation. The NL-VO set is defined as follows:

[0127] (18)

[0128] Among them, is a set of non - linear speed obstacles, and respectively represent the current positions of the stand - on vessel and the give - way vessel, describes all the collision - risk positions that the give - way vessel may be in at time . denotes Minkowski addition, which is used to calculate the prohibited area around the stand - on vessel (i.e., the area that the give - way vessel should not enter).

[0129] From Figure 3 , it can be seen that in the speed space, the speed space of the stand - on vessel is divided into multiple regions for quantifying the current collision - avoidance decision risk. The starting point O represents the position of the stand - on vessel when its speed is zero and serves as the center of the speed space. All possible speed vectors start from this point. The current speed state is represented by a vector, where the length represents the speed magnitude and the direction represents the speed direction. The speed space is divided into a forward speed safety zone (S1), a non - linear speed obstacle zone (S2), and a rear speed safety zone (S3). By calculating the distances from and to the boundaries and the starting point of each region, the risk of the current strategy can be measured, providing a basis for real - time decision - making.

[0130] First, calculate the distance between the current speed of the stand - on vessel and the non - linear speed - obstacle boundary :

[0131] (19)

[0132] Among them, represents the speed - obstacle boundary point corresponding to the speed vector of the stand - on vessel at time , which is defined as:

[0133] (20)

[0134] The maximum distance is defined as:

[0135] (21)

[0136] The normalized risk metric is expressed as:

[0137] (22)

[0138] At the same time, calculate the current speed Distance to the vertex of the collision triangle : :

[0139] (23)

[0140] The maximum distance is defined as:

[0141] (24)

[0142] The corresponding normalized risk measure is:

[0143] (25)

[0144] Since a change in speed is usually accompanied by a change in heading, it is necessary to quantify the impact of the change in heading on the collision risk. The heading of a straight-going ship is expressed by the following formula:

[0145] (26)

[0146] where is the component of the speed of the straight-going ship in the axis direction, is the component of the speed of the straight-going ship in the axis direction. This formula represents the angle between the current speed vector of the straight-going ship and the vertical axis, representing the current heading.

[0147] Based on this, the deviation of the heading is defined as:

[0148] (27)

[0149] where is the heading represented by the optimal speed, is the angle between the current speed of the straight-going ship and the optimal speed.

[0150] By combining these factors, a comprehensive formula for quantifying the heading risk can be constructed:

[0151] (28)

[0152] where is the angle between the current speed of the straight-going ship and the maximum speed change on the left. is the angle between the current speed of the straight-going ship and the maximum speed change on the right. is the weight factor, which is used to balance the contribution of the angular deviation to the overall risk. This formula provides a comprehensive index for the decision-making process to measure the impact of the course deviation on the collision risk.

[0153] Based on the quantization methods of speed and course, a comprehensive risk quantization formula is established, and the risk factors of speed and course are combined through a weighting method:

[0154] (29)

[0155] where represents the distance between the current speed and the boundary of the non-linear speed obstacle area, reflecting the danger in space. represents the distance between the current speed and the starting point of the collision triangle reflecting the urgency in time. represents the risk quantization of the course, combining multiple factors of course change and safety distance. are respectively 、 and are the weighting coefficients of, used to balance the contributions of speed and course to the overall risk.

[0156] Therefore, the strategy safety objective function and the economic objective function are respectively:

[0157] (30)

[0158] (31)

[0159] where represents the speed of the stand-on vessel, represents the turning radius in the collision avoidance operation, and the decision variable , , represent the speed change ratio, the course change amount, and the avoidance duration respectively; rad() represents radians.

[0160] In order to further verify the adaptability of the model in complex encounter situations, this embodiment designs a multi-ship encounter scenario. In this scenario, the stand-on ship (OS) serves as the stand-on ship, and the give-way ship 1 (TS1) and the give-way ship 2 (TS2) serve as the give-way ships. Three situations are considered: both TS1 and TS2 cooperate, TS1 does not cooperate, and neither TS1 nor TS2 cooperate. The research objective is to evaluate the situation awareness and decision-making optimization capabilities of the stand-on ship in multi-ship encounters under different cooperation conditions. The initial conditions and parameters are shown in Table 2.

[0161] Table 2 Navigation parameters of the multi-ship scenario

[0162] In scenario 2.1, as the give-way vessels, TS1 and TS2 executed collision avoidance measures in accordance with the International Regulations for Preventing Collisions at Sea. The trajectories of the three vessels are as follows Figure 4 shown. TS1 and TS2 avoided collision with the through vessel by reducing speed and turning right.

[0163] As Figure 5 shown in and Table 2, TS2 initiated collision avoidance at 516 s, but due to reaction delay, its collision avoidance measures did not significantly reduce the collision risk (see Figure 6 ). When the collision risk reached the set threshold at 520 s, the through vessel immediately executed a 20-degree right turn and reduced speed to 7.4 m / s, performing collision avoidance for 440 s. TS1 initiated collision avoidance at 705 s, but due to the continuous avoidance of the through vessel, the collision risk did not increase further. At 805 s, the through vessel entered a new decision cycle and executed a 20-degree left turn for 240 s. Finally, the collision risk with TS1 and TS2 was reduced to zero. The through vessel resumed its original course at 1045 s and fully restored its initial track at 1320 s.

[0164] Figure 7 shows the non-linear velocity obstacle (NL-VO) analysis of the through vessel with TS1 and TS2. By comparing the NL-VO analysis at each decision moment, the figure shows how the through vessel gradually moved away from the collision risk area after executing the collision avoidance action. Figure 7 As shown in Figure (a) in, at 520 s, the velocity point of the through vessel was located within the NL-VO, indicating the existence of a collision risk. Figure 7 As shown in Figures (b) and (c) in, after two decision moments, the velocity point of the through vessel gradually moved away from the NL-VO boundary, successfully reducing the collision risk. Finally, Figure 7 Figures (d) and (e) in confirmed the effectiveness of the collision avoidance decision after the through vessel resumed its course.

[0165] In scenario 2.2, TS1 maintained a constant speed and course, Figure 8 shows the trajectory of the through vessel, which was obtained through give-way vessel intention reasoning and decision optimization. The relevant navigation data are shown in Figure 9 and Table 2. As a cooperative give-way vessel, TS2 initiated collision avoidance at 516 s by turning right 30 degrees and reducing speed to 6.16 m / s. However, due to reaction delay, its collision avoidance measures did not significantly reduce the collision risk (see Figure 10 ).

[0166] When the collision risk reached the set threshold at 520 seconds, the straight-going ship executed a 30-degree right turn and decelerated to 7.4 m / s, and carried out collision avoidance for 245 seconds. At 705 seconds, TS1 started collision avoidance, but due to the continuous avoidance of the straight-going ship, the collision risk did not increase. At 805 seconds, the straight-going ship entered a new decision cycle, executed a 20-degree left turn, and carried out collision avoidance for 240 seconds. The collision risks with TS1 and TS2 dropped to zero. At 1045 seconds, the straight-going ship began to resume its course and completely restored its original track at 1320 seconds.

[0167] Figure 11 shows the variation of the collision risk over time, demonstrating the collision avoidance actions of the straight-going ship. Figure 11 The NL-VO analysis in Figure 11 confirmed the effectiveness of the strategy: as shown in Figure (a) in Figure 11 , at 520 seconds, the speed point of the straight-going ship was located within the NL-VO, indicating the existence of a collision risk; Figure 11 as shown in Figures (b) and (c) in

[0168] , after the first decision, the speed point gradually moved away from the NL-VO boundary, reducing the risk; Figure 12 as shown in Figures (d) and (e) in

[0169] As shown in Figure 13 , at 315 seconds, the collision risk between OS and TS2 reached the threshold (see Figure 14 ), and the straight-going ship initiated the first collision avoidance, turning 30 degrees to the right and reducing the speed to 5.76 m / s for 340 seconds. At 720 seconds, the collision risk between OS and TS1 reached the threshold again, and the straight-going ship initiated the second collision avoidance, turning 20 degrees to the left for 340 seconds. At 1060 seconds, OS initiated the third collision avoidance, turning 10 degrees to the left for 375 seconds. By 1810 seconds, OS had resolved the collision risks with TS1 and TS2 and restored its original course.

[0170] Figure 15 shows the NL-VO analysis of the collision avoidance decision of OS. Figure 15 As shown in Figure (a) in Figure 15 , at 520 seconds, the speed point of OS was located within the NL-VOs of TS1 and TS2, indicating a relatively high collision risk. Figure 15As shown in Figure (c), at 754 seconds, the speed point of the OS is located at the NL-VO edge of TS1 and outside the NL-VO of TS2, successfully reducing the risk with TS1. Figure 15 As shown in Figure (d), after the third collision avoidance, the speed point of the OS has moved out of the NL-VO of TS2 but is still within the NL-VO of TS1, indicating that further collision avoidance is required. Finally, Figure 15 Figures (e) and (f) confirm that the speed point of the OS has completely moved out of the two NL-VOs, achieving complete collision avoidance.

[0171] Figure 16 The comparison effect of the collision decision risk and route loss corresponding to before and after the first avoidance opportunity of each experiment is shown. It can be seen that the avoidance opportunity determined according to the present invention is the optimal avoidance opportunity. Making the route loss increase too early and causing collision risk too late.

[0172] The content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

Claims

1. A collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention, characterized in that, Including: Regarding the relative motion of the stand-on vessel and the give-way vessel, dividing the collision risk level according to the collision risk index of the stand-on vessel and the give-way vessel; Constructing a cooperation relationship matrix in the global encounter situation within the observable range of the stand-on vessel, judging whether the spatio-temporal effect difference between the predicted state and the actual observed state of each give-way vessel from the perspective of the stand-on vessel exceeds the set threshold. If so, determining that the intention of the give-way vessel is non-cooperative, adjusting the cooperation relationship matrix, and judging whether the collision risk index exceeds the set threshold corresponding to the collision risk level. If so, triggering the stand-on vessel to execute an avoidance action; After triggering the avoidance action, using the non-linear velocity obstacle method to calculate the collision avoidance decision risk, constructing a safety objective function, constructing an economic objective function based on the speed and turning radius, and optimizing the safety objective function and the economic objective function through the NSGA-II algorithm to obtain the optimal collision avoidance decision.

2. The collision avoidance decision-making method for a stand-on vessel according to claim 1 based on the inference of the give-way vessel's intention, characterized in that, The collision risk index is obtained by normalizing the degree of domain intrusion and the remaining time of domain intrusion. The collision risk levels of the stand-on vessel and the give-way vessel are divided into four stages: cautious, warning, dangerous, and extremely dangerous according to the value of the collision risk index. The collision risk index CRI is determined by the following formula: ； Where, m is the collision risk adjustment factor for the degree of domain intrusion; n is the collision risk adjustment factor for the remaining time of domain intrusion; DDV is the degree of domain intrusion; TDV is the remaining time of domain intrusion.

3. The collision avoidance decision-making method for a stand-on vessel according to claim 2, which is characterized in that Using the eccentric ellipse domain model to calculate the degree of domain intrusion and the remaining time of domain intrusion respectively. The method is as follows: setting the major axis length, minor axis length and corresponding offset of the eccentric ellipse domain according to the hull size and speed of the stand-on vessel; substituting the relative position of the give-way vessel relative to the stand-on vessel into the eccentric ellipse domain equation. If the intrusion judgment formula is satisfied, it indicates that the give-way vessel intrudes into this domain; solving the relative motion equation of the give-way vessel and the stand-on vessel to obtain the time root when the give-way vessel enters or leaves the eccentric ellipse domain, and thereby determining the degree of domain intrusion and the remaining time of domain intrusion, which are used to characterize the depth and urgency of the collision risk.

4. A collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention according to claim 1, characterized in that, The spatio-temporal effect difference between the predicted state and the actual observed state of the give-way vessel from the perspective of the stand-on vessel is obtained through the following steps: constructing the actual observed state sets of the stand-on vessel and the give-way vessel, generating the predicted state set of the give-way vessel at a certain moment after a period of time based on the decision-making and observation information after a period of time, and obtaining the predicted space effect coefficient and the predicted time effect coefficient through data processing; obtaining the new actual observed state set of the give-way vessel after a period of time, and obtaining the actual observed space effect coefficient and the actual observed time effect coefficient through data processing; obtaining the spatio-temporal effect corresponding to the predicted state and the spatio-temporal effect corresponding to the actual observed state through the BPA function for the predicted space effect coefficient and the predicted time effect coefficient, the actual observed space effect coefficient and the actual observed time effect coefficient, and fusing the spatio-temporal effect corresponding to the predicted state and the spatio-temporal effect corresponding to the actual observed state.

5. A collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention according to claim 4, characterized in that, The spatio-temporal effect difference between the predicted state and the actual observed state of the give-way vessel from the perspective of the stand-on vessel after a period of time is expressed as: ； ； ； wherein, represents the current moment a period of time later the confidence level of the predicted state of the give-way vessel supporting cooperation from the perspective of the stand-on vessel; represents the current moment a period of time later the confidence level of the predicted state of the give-way vessel not supporting cooperation from the perspective of the stand-on vessel; represents the current moment a period of time later the degree to which the given collision avoidance behavior of the predicted state of the give-way vessel from the perspective of the stand-on vessel is recognized as supported or not supported; represents the current moment a period of time later the confidence level of the actual observed state of the give-way vessel supporting cooperation from the perspective of the stand-on vessel; represents the current moment a period of time later the confidence level of the actual observed state of the give-way vessel not supporting cooperation from the perspective of the stand-on vessel; represents the current moment a period of time later the degree to which the given collision avoidance behavior of the actual observed state of the give-way vessel from the perspective of the stand-on vessel is recognized as supported or not supported; represents the current moment a period of time later the spatio-temporal effect corresponding to the actual observed state of the give-way vessel from the perspective of the stand-on vessel; represents the current moment a period of time later the spatio-temporal effect corresponding to the predicted state of the give-way vessel from the perspective of the stand-on vessel; represents the current moment a period of time later the spatio-temporal effect difference between the predicted state and the actual observed state of the give-way vessel from the perspective of the stand-on vessel.

6. The collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention according to claim 1, characterized in that Cooperation relationship matrix Determined by the following formula: ； Among them, They are respectively the cooperation relationships between the 1st, 2nd, 3rd... nth give-way vessels and the stand-on vessel from the perspective of the stand-on vessel. The value 1 represents cooperation, and the value 0 represents non-cooperation.

7. A collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention according to claim 1, wherein Building a safety objective function includes the following steps: mapping the relative position, speed, and course of the stand-on vessel and the give-way vessel to the speed space, quantifying the spatial danger by judging the distance between the speed vector of the stand-on vessel and the boundary of the non-linear speed obstacle area, and quantifying the time urgency by the distance from the collision triangle vertex; calculating the deviation between the current course of the stand-on vessel and the optimal speed course, and weighting and quantifying the course deviation risk in combination with the included angle of the maximum variable speed directions on the left and right; integrating and quantifying the spatial danger, time urgency, and course deviation risk to obtain the safety objective function.

8. A collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention according to claim 7, characterized in that The safety objective function is determined by the following formula: Safety objective function f1: ； Among them, represents the distance between the current speed and the boundary of the non-linear speed obstacle area; represents the distance between the current speed and the starting point of the collision triangle ; represents the risk quantification of the heading; are respectively , and weighting coefficients of.

9. The collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention according to claim 1, characterized in that The economic objective function is determined by the following formula: Economic objective function f2: ； Among them, represents the ship's speed; represents the turning radius in collision avoidance operations; decision variable , , respectively represent the speed change ratio, the course change amount, and the avoidance duration; rad() represents radians.

10. A collision avoidance decision-making method for a stand-on vessel based on the inference of the give-way vessel's intention according to claim 1, characterized in that After triggering the avoidance action, it also includes: defining the collision avoidance decision as three main decision variables: the speed change ratio, the course change amount, and the avoidance duration, and using the Nomoto model to evaluate the response of the stand-on vessel to these decision variables.

Images