Ship avoidance path generation method and device based on data driving

Abstract

The invention provides a ship avoidance path generation method and device based on data driving, and the method comprises the steps: extracting a ship historical meeting scene containing an avoidance behavior from historical AIS data, and recognizing the historical scene features and historical avoidance features of the ship historical meeting scene, constructing a database based on the historical scene features and the historical avoidance features; acquiring real-time scene features of the real-time navigation scene, performing similarity matching on the real-time scene features and historical scene features in a database, determining at least one target historical scene, and determining target historical avoidance features corresponding to the target historical scene; generating at least one candidate avoidance path for the real-time navigation scene based on the historical avoidance features; and performing maneuverability and safety evaluation on the at least one candidate avoidance path, and determining an optimal avoidance path. According to the method, the reliability, the safety and the interpretability of the generated avoidance path are improved, and a key technical support is provided for the development of intelligent shipping.

Description

Technical Field

[0001] This invention relates to the field of ship collision avoidance technology, specifically to a data-driven method and apparatus for generating ship collision avoidance paths. Background Technology

[0002] Maritime traffic safety is a critical concern for the international community, as more than four-fifths of global trade is conducted by sea, with collisions being the primary type of maritime accident or incident. Research on ship collision avoidance is considered a key and effective means of reducing such accidents, aiming to answer what actions should be taken to prevent collisions when a ship is in distress.

[0003] There are two existing methods for preventing ship collisions. One method is to solve for collision avoidance paths based on mathematical models such as ship dynamics / dynamic models. However, mathematical models cannot fully simulate the complex hydrodynamics, wind, waves, currents, and ship characteristics in the real marine environment, resulting in low accuracy of the determined collision avoidance paths and poor adaptability to actual scenarios. The other method is based on deep learning for collision avoidance path determination. However, the decision-making process of deep learning is opaque and has poor interpretability.

[0004] Therefore, there is an urgent need for a data-driven method and apparatus for generating ship collision avoidance paths that can both be data-driven and intelligent, and ensure that the generated decisions conform to the physical characteristics of the ship, thereby improving the accuracy and interpretability of ship collision avoidance paths. Summary of the Invention

[0005] In view of this, it is necessary to provide a data-driven method and apparatus for generating ship collision avoidance paths, in order to solve the technical problems in the prior art that the reliance on mathematical models leads to poor accuracy of ship collision avoidance paths and the reliance on deep learning leads to poor interpretability of the ship collision avoidance path decision-making process.

[0006] To address the aforementioned technical problems, in a first aspect, the present invention provides a data-driven method for generating ship avoidance paths, comprising: Extract historical encounter scenarios of ships containing avoidance behavior from historical AIS data, identify historical scene features and historical avoidance features of the historical encounter scenarios, and construct a database based on the historical scene features and the historical avoidance features; The real-time scene features of the real-time navigation scene are obtained, and the real-time scene features are matched with the historical scene features in the database to determine at least one target historical scene and to determine the target historical avoidance features corresponding to the target historical scene. Based on the historical avoidance features, at least one candidate avoidance path is generated for the real-time navigation scenario; Maneuverability and safety are evaluated for the at least one candidate avoidance path to determine the optimal avoidance path.

[0007] In one possible implementation, the historical scenario features include the encounter situation between the vessel performing the avoidance maneuver and the target vessel being avoided, and other features, including the nearest encounter distance, the nearest encounter time, relative distance, relative speed, and relative bearing.

[0008] In one possible implementation, identifying the encounter situation of the ship's historical encounter scenarios includes: Determine the encounter situation of the target vessel based on its relative bearing to the target vessel. Determine the encounter situation of the target ship based on the relative bearing of the ship to the target ship; The encounter situation is obtained by combining the encounter situation of the current vessel and the encounter situation of the target vessel.

[0009] In one possible implementation, the encounter situation of the vessel includes categories A, B, C, D, E, F, and G, where the relative bearing of the target vessel relative to the vessel in category A is […]. The relative bearing of the target vessel (Category B) to this vessel is [ The relative bearing of the target vessel (Category C) to this vessel is [ The relative bearing of the target vessel (Category D) to this vessel is [ The relative bearing of the target vessel of category E to this vessel is [ The relative bearing of the target vessel of category F with respect to this vessel is [ The relative bearing of the target vessel (class G) to this vessel is [ ).

[0010] In one possible implementation, the historical avoidance features include the starting point of the ship's avoidance behavior, the ending point of the avoidance behavior, the point where the turning rate is at an extreme value, and the trajectory point when the turning rate is zero.

[0011] In one possible implementation, the historical avoidance feature takes the starting point of the avoidance behavior as the pole and uses radial distance and polar angle to characterize the ending point of the avoidance behavior, the trajectory point when the steering rate is at its extreme value, and the trajectory point when the steering rate is zero.

[0012] In one possible implementation, the real-time scene features include real-time encounter situations and other real-time features; then, the step of performing similarity matching between the real-time scene features and historical scene features in the database to determine at least one target historical scene includes: Based on the real-time encounter situation, at least one candidate historical encounter scenario with the same encounter situation as the real-time encounter situation is selected from the database. The at least one target historical scenario is determined from the at least one candidate historical encounter scenario based on other features of each candidate historical encounter scenario and the other real-time features.

[0013] In one possible implementation, determining the at least one target historical scenario among the at least one candidate historical encounter scenario based on other features of each of the candidate historical encounter scenarios and the other real-time features includes: Calculate the first cumulative distribution function value of the other real-time features and the second cumulative distribution function value of the other features corresponding to each of the target historical scenes; Determine whether the absolute difference between the first cumulative distribution function value and the second cumulative distribution function value is less than a preset difference; If the value is less than the target historical scenario, then the candidate historical encounter scenario will be used as the target historical scenario.

[0014] In one possible implementation, the step of performing a maneuverability and safety assessment on the at least one candidate avoidance path to determine the optimal avoidance path includes: The candidate avoidance path is tracked and simulated using a ship maneuvering motion model, a line-of-sight guidance model, and a PID controller to obtain the tracking path and determine the mean absolute error between the tracking path and the candidate avoidance path. Predict the target ship's position at a future time and calculate the minimum distance between the tracking path and the target ship; The candidate avoidance path that has the minimum distance greater than the preset safety distance and the minimum average absolute error is selected as the optimal avoidance path.

[0015] Secondly, the present invention also provides a data-driven ship avoidance path generation device, comprising: The database construction unit is used to extract historical encounter scenarios of ships containing avoidance behavior from historical AIS data, identify historical scene features and historical avoidance features of the historical encounter scenarios, and construct a database based on the historical scene features and the historical avoidance features. The scene matching unit is used to acquire real-time scene features of a real-time navigation scene, perform similarity matching between the real-time scene features and historical scene features in the database, determine at least one target historical scene, and determine the target historical avoidance features corresponding to the target historical scene. A candidate avoidance path generation unit is used to generate at least one candidate avoidance path for the real-time navigation scenario based on the historical avoidance features. The obstacle avoidance path optimization unit is used to perform maneuverability and safety assessments on the at least one candidate obstacle avoidance path and determine the optimal obstacle avoidance path.

[0016] The beneficial effects of this invention are as follows: The data-driven ship collision avoidance path generation method provided by this invention extracts historical ship encounter scenarios containing collision avoidance behaviors from historical AIS data, constructs a database based on these scenarios, and makes collision avoidance path decisions based on the database. This ensures that the collision avoidance path is no longer simply the result of mathematical optimization, but rather utilizes the collective experience of human drivers in similar scenarios. This overcomes the problem that traditional mathematical model methods do not conform to actual navigation, making the generated collision avoidance path more consistent with maritime practice, thereby improving the accuracy and scenario adaptability of the generated ship collision avoidance path. Secondly, this invention determines candidate collision avoidance paths through feature matching, clearly indicating that the collision avoidance path is based on a similar historical scenario. This case-driven explanation method has strong interpretability.

[0017] Furthermore, after determining the candidate avoidance paths, the present invention determines the optimal avoidance path through maneuverability and safety assessment, ensuring that the final determined avoidance path is suitable for the vessel's maneuverability and can maintain a safe distance from other vessels, thereby further improving the reliability and safety of the generated avoidance path.

[0018] In summary, this invention combines data-driven artificial intelligence with professional physical models in the maritime field to generate intelligent, reliable, safe, and feasible ship avoidance paths. It fundamentally solves the shortcomings of traditional methods in terms of practicality, interpretability, and reliability, and provides key technical support for the development of intelligent shipping. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic flowchart of an embodiment of the data-driven ship avoidance path generation method provided by the present invention; Figure 2 A schematic flowchart of an embodiment of the present invention for identifying encounter situations in historical ship encounter scenarios; Figure 3 A schematic diagram of an embodiment of the encounter situation provided by the present invention; Figure 4 A schematic diagram of an embodiment of a historical encounter scenario involving avoidance behavior provided by the present invention; Figure 5 A schematic diagram illustrating an embodiment of the ship's course and turning rate provided by the present invention; Figure 6 For the present invention Figure 1 A schematic diagram of an embodiment of S104; Figure 7 For the present invention Figure 6 A schematic diagram of an embodiment of S602; Figure 8 A schematic diagram of an embodiment of the relative distance similarity judgment provided by the present invention; Figure 9 For the present invention Figure 1 A schematic diagram of an embodiment of S104; Figure 10 This is a schematic diagram of an embodiment of the data-driven ship avoidance path generation device provided by the present invention. Detailed Implementation

[0021] 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 a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] It should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Furthermore, those skilled in the art, guided by the content of this invention, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0023] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0024] This invention provides a data-driven method and apparatus for generating ship avoidance paths, which will be described below.

[0025] Figure 1 A schematic flowchart of an embodiment of the data-driven ship avoidance path generation method provided by the present invention is shown below. Figure 1 As shown, the data-driven method for generating ship avoidance paths includes: S101. Extract historical encounter scenarios of ships containing avoidance behavior from historical AIS data, identify historical scene features and historical avoidance features of historical encounter scenarios, and construct a database based on historical scene features and historical avoidance features.

[0026] Specifically, historical AIS data includes ship tracks. Historical encounter scenarios of ships can be represented as ,in, Indicates the trajectory of evasive action. Indicates the trajectory of the vessel being avoided, t i and t j These are the start and end times of the avoidance action, respectively.

[0027] The extraction process for historical vessel encounter scenarios is as follows: First, historical AIS data is preprocessed, such as removing outliers and interpolating to compensate for missing values. Second, based on collision risk indicators (such as the nearest encounter distance and the nearest encounter time), it is determined whether there is a collision risk between the two vessels. If so, the heading data of the vessel under collision risk (OS) is analyzed to identify obvious and continuous turning maneuvers (i.e., avoidance maneuvers). Then, it is confirmed that the turning behavior is an avoidance behavior against a specific target vessel (TS), rather than for other reasons (such as berthing / departure, turning points, etc.). Finally, the start and end times of the avoidance behavior are determined to obtain the historical vessel encounter scenario.

[0028] S102. Obtain the real-time scene features of the real-time navigation scene, perform similarity matching between the real-time scene features and the historical scene features in the database, determine at least one target historical scene, and determine the target historical avoidance features corresponding to the target historical scene. S103. Generate at least one candidate avoidance path for the real-time navigation scenario based on historical avoidance features.

[0029] Specifically, the real-time position of the ship in the real-time navigation scenario is determined, and the sequence of avoidance feature points is obtained from the historical scenario. Then, the path point of the candidate avoidance path = the real-time position of the ship + the historical displacement vector determined based on the avoidance feature points.

[0030] S104. Conduct a maneuverability and safety assessment on at least one candidate avoidance path to determine the optimal avoidance path.

[0031] It should be noted that the ship avoidance in the embodiments of the present invention only involves a change in course and does not involve a change in speed.

[0032] It should be understood that the data-driven ship collision avoidance path generation method in this embodiment of the invention can be implemented in any device based on a data-driven ship collision avoidance path generation method, such as a ship or shore control center based on a data-driven ship collision avoidance path generation method. Specifically, the data-driven ship collision avoidance path generation method is stored in the aforementioned device as a pre-programmed program. When the device is started, the program is invoked, and the data-driven ship collision avoidance path generation method is implemented.

[0033] Compared with existing technologies, the data-driven ship collision avoidance path generation method provided in this invention extracts historical ship encounter scenarios containing collision avoidance behaviors from historical AIS data, constructs a database based on these scenarios, and makes collision avoidance path decisions based on the database. This makes the collision avoidance path no longer a simple result of mathematical optimization, but rather utilizes the collective experience of human pilots in similar scenarios. This overcomes the problem that traditional mathematical model methods do not conform to actual navigation, making the generated collision avoidance path more consistent with maritime practice, thereby improving the accuracy and scenario adaptability of the generated ship collision avoidance path. Secondly, this invention determines candidate collision avoidance paths through feature matching, clearly indicating that the collision avoidance path is based on a similar historical scenario. This case-driven explanation method has strong interpretability.

[0034] Furthermore, in this embodiment of the invention, after determining the candidate avoidance paths, the optimal avoidance path is determined through maneuverability and safety assessment, ensuring that the final determined avoidance path is suitable for the vessel's maneuverability and can maintain a safe distance from other vessels, thereby further improving the reliability and safety of the generated avoidance path.

[0035] In summary, the embodiments of the present invention combine data-driven artificial intelligence with professional physical models in the field of navigation to generate intelligent, reliable, safe and feasible ship avoidance paths. This fundamentally solves the shortcomings of traditional methods in terms of practicality, interpretability and reliability, and provides key technical support for the development of intelligent shipping.

[0036] To improve the accuracy of scene matching, in a specific embodiment of the present invention, the historical scene features include the encounter situation between the ship that performs the avoidance behavior and the target ship that is being avoided, and other features, including the nearest encounter distance, the nearest encounter time, the relative distance, the relative speed, and the relative bearing.

[0037] Specifically, the formulas for calculating the nearest encounter distance (DCPA) and the nearest encounter time (TCPA) are as follows:

[0038] In the formula, d r It is a relative distance; v r It is the relative velocity; c b It is the relative azimuth angle; c r This refers to the relative heading.

[0039] The formulas for calculating relative distance, relative velocity, and relative azimuth are as follows:

[0040] In the formula, ( x os , y os ( ) represents the coordinates of the ship's trajectory points; x ts ,y ts ( ) represents the coordinates of the target ship's trajectory points; ( ) represents the velocity components of the ship's trajectory point along the X and Y axes of the coordinate system; () represents the velocity components of the target ship's trajectory point along the X and Y axes of the coordinate system. c o This is the heading of the ship at the track point.

[0041] The International Regulations for Preventing Collisions at Sea (ICMLS) disclose three encounter scenarios: head-on, crossing, and overtaking. To further improve the accuracy of scenario matching, in some embodiments of this invention, such as... Figure 2 As shown, the encounter situations for identifying historical ship encounter scenarios include: S201. Determine the encounter situation of the target vessel based on its relative bearing to the target vessel. S202. Determine the encounter situation of the target ship based on the relative bearing of the ship to the target ship. S203. Combine the encounter situation of this vessel with the encounter situation of the target vessel to obtain the encounter situation; For example, if the encounter situation of our ship is A and the encounter situation of the target ship is B, the encounter scenario is AB. That is, the combination of the encounter situation of our ship and the encounter situation of the target ship is the encounter scenario.

[0042] The present invention combines the encounter situation of the ship and the encounter situation of the target ship as the encounter scenario. Compared with the three encounter scenarios in the prior art, it increases the types of encounter scenarios, improves the refinement of the encounter scenarios, and thus improves the scene matching accuracy.

[0043] In specific embodiments of the present invention, such as Figure 3 As shown, the encounter situations of this vessel include categories A, B, C, D, E, F, and G. The relative bearing of the target vessel in category A is […]. The relative bearing of the target vessel (Category B) to this vessel is [ The relative bearing of the target vessel (Category C) to this vessel is [ The relative bearing of the target vessel (Category D) to this vessel is [ The relative bearing of the target vessel of category E to this vessel is [ The relative bearing of the target vessel of category F with respect to this vessel is [ The relative bearing of the target vessel (class G) to this vessel is [ ).

[0044] Similarly, the encounter situations of the target vessel also include the seven categories mentioned above. Figure 3 The bottom left corner represents our ship, and the top right corner represents the target ship. Figure 3 The situation encountered is BG.

[0045] This invention, by subdividing the encounter situation of the vessel and the target vessel, can increase the number of encounter scenarios, thereby refining the historical encounter scenarios of ships and improving the precision of the determined avoidance paths.

[0046] To comprehensively describe the heading avoidance behavior of a ship, in a specific embodiment of the present invention, the historical avoidance features include the starting point of the ship's avoidance behavior, the ending point of the avoidance behavior, the point where the turning rate is at its extreme value, and the trajectory point when the turning rate is zero.

[0047] Among these parameters, the rate of change of course (ROT) is a key indicator reflecting the rate of change of a ship's course. When ROT equals zero, it means that the rate of change of course has dropped to zero. When ROT reaches its extreme value, the ship's rate of change of course is at its fastest. These parameters effectively reflect the ship's turning characteristics. Therefore, introducing points with extreme rates of change of course and trajectory points with zero rates of change of course into historical collision avoidance data can achieve a comprehensive description of the ship's collision avoidance behavior and improve the accuracy of subsequent similarity matching.

[0048] In one specific embodiment of the present invention, Figure 4 For historical encounter scenarios involving yielding behavior, the horizontal axis represents longitude and the vertical axis represents latitude, such as... Figure 4 As shown, the blue and black curves represent the tracks of the vessel performing the evasive maneuver and the target vessel being evaded, respectively. The vessel's track is represented as... The target ship's track is represented as . , . Figure 5The diagram at the top center shows how the ship's course changes over time. Figure 5 The lower middle chart represents ROT data. Figure 5 The black, red, and blue dots represent the start time of the avoidance maneuver, the end time of the avoidance maneuver, the moment when the turning rate reaches its extreme value, and the moment when the turning rate reaches zero, respectively. These moments can be used to determine... Figure 4 The trajectory points in the data, namely the starting point of the avoidance behavior, the ending point of the avoidance behavior, the point where the turning rate is at its extreme value, and the trajectory point when the turning rate is zero, constitute the historical avoidance characteristics.

[0049] Specifically, historical avoidance characteristics can be represented as:

[0050] In the formula, t s+1 To t s+n These are the moments when the steering rate is at its extreme value and when the steering rate is zero.

[0051] The historical scenario takes place at a specific location in the world coordinate system (e.g., 122 degrees east longitude, 31 degrees north latitude). If the latitude and longitude points of the avoidance trajectory are recorded directly, then this avoidance path is only applicable to this specific location and cannot be directly applied to a real-time scenario occurring in a completely different geographical location. Furthermore, even if the two scenarios are geographically close, the initial headings of the ships may be completely different.

[0052] Considering the aforementioned technical problems, in order to convert historical obstacle avoidance features in specific scenarios into abstract features applicable to real-time scenarios, in a specific embodiment of the present invention, after obtaining historical obstacle avoidance features based on position and heading representations, it is necessary to transform them to possess translational invariance and rotational invariance. Specifically, the historical obstacle avoidance features take the starting point of the obstacle avoidance behavior as the pole, and use radial distance and polar angle to characterize the ending point of the obstacle avoidance behavior, the trajectory point when the turning rate is at its extreme value, and the trajectory point when the turning rate is zero.

[0053] Specifically, the characteristics of historical avoidance are:

[0054]

[0055] In the formula, Let be the radial distance of the i-th historical avoidance feature; Let be the polar angle of the i-th historical avoidance feature; To avoid the coordinates of the starting point of the action; Let be the coordinates of the i-th historical avoidance feature.

[0056] The embodiments of the present invention characterize historical avoidance features by radial distance and polar angle, decoupling and abstracting historical avoidance features from specific spatiotemporal contexts, so that they can be transferred to new real-time scenarios, thereby improving the applicability of historical avoidance features.

[0057] When there are many historical ship encounter scenarios in the database, directly performing similarity matching based on historical scenario features and historical avoidance features would be computationally intensive, making it difficult to meet the requirements of real-time ship avoidance path decision-making. To address this technical problem, in some embodiments of this invention, real-time scenario features include real-time encounter situations and other real-time features; for example... Figure 6 As shown, step S104 includes: S601. Based on the real-time encounter situation, select at least one candidate historical encounter scenario from the database whose encounter situation is the same as the real-time encounter situation. S602. Based on other features of each candidate historical encounter scenario and other real-time features, determine at least one target historical scenario in at least one candidate historical encounter scenario.

[0058] This invention first performs a preliminary screening of historical encounter scenarios in the database based on the encounter situation, obtaining candidate historical encounter scenarios. Subsequent similarity matching is then performed only within these candidate scenarios, significantly reducing computational load and meeting the real-time requirements of ship avoidance paths. For example, assuming the database contains 10,000 historical scenarios evenly distributed across 7 encounter situations, after the first screening, each subset contains approximately 10,000 / 7 ≈ 1429 scenarios. Subsequent matching only needs to be performed on these 1429 scenarios, instead of all 10,000. This directly reduces computational load by approximately 86%, which is crucial for ensuring the real-time nature of collision avoidance decisions.

[0059] Furthermore, the experience of avoiding a meeting situation (usually a right turn) is fundamentally different from the experience of avoiding an overtaking situation (the overtaking vessel must give way). Matching a meeting scenario with an overtaking scenario, even if their DCPA, TCPA, and other values ​​happen to be similar, will result in an incorrect and dangerous avoidance path. This embodiment of the invention, by determining the target historical scenario within candidate historical meeting scenarios with the same meeting situation, ensures that the matched scenarios have the same avoidance rule basis, thus avoiding fundamental errors.

[0060] In specific embodiments of the present invention, such as Figure 7 As shown, step S602 includes: S701. Calculate the first cumulative distribution function value of other real-time features and the second cumulative distribution function value of other features corresponding to each target's historical scene.

[0061] Specifically, calculate the two cumulative distribution function values ​​for the nearest encounter distance, the nearest encounter time, the relative distance, the relative velocity, and the relative azimuth angle.

[0062] S702. Determine whether the absolute difference between the first cumulative distribution function value and the second cumulative distribution function value is less than a preset difference.

[0063] Specifically, the absolute difference needs to be calculated for each of the following: nearest encounter distance, nearest encounter time, relative distance, relative speed, and relative azimuth.

[0064] S703. If it is less than, then the candidate historical encounter scene will be used as the target historical scene.

[0065] Similarly, the candidate historical encounter scene will be selected as the target historical scene only when the differences in the five dimensions of the nearest encounter distance, the nearest encounter time, the relative distance, the relative speed, and the relative azimuth angle are all less than the preset difference.

[0066] In a specific embodiment of the present invention, taking the similarity judgment of relative distance as an example, such as... Figure 8 As shown, the blue line represents the cumulative distribution function (CDF) curve of the relative distance feature, and the blue dots correspond to the CDF values ​​of the relative distance feature in the real-time scene. The red dots represent similarity probability thresholds. The calculated CDF critical point. When the... The relative distance between historical scenarios When the scene falls within the shaded area, it is considered to have a high degree of similarity to the real-time scene in terms of relative distance features.

[0067] Due to differences in vessel type and load capacity, these vessels exhibit varying maneuverability in real-time and matched scenarios. This means that the generated avoidance paths may not always be suitable for the vessel's maneuverability in real-time scenarios. Therefore, after generating multiple candidate avoidance paths, maneuverability and safety assessments are still required. Specifically, such as... Figure 9 As shown, step S104 includes: S901. Using a ship maneuvering motion model, line-of-sight (LOS) model, and PID controller, the candidate avoidance path is tracked and simulated to obtain the tracking path and determine the mean absolute error between the tracking path and the candidate avoidance path.

[0068] Specifically, the target heading deviation is calculated by using the LOS model in combination with the headings of the starting point and the candidate avoidance path. Then, a PID controller is used to establish a heading controller. The set desired heading deviation is input into the PID controller, and the required rudder angle is output. Finally, the ship maneuvering motion model is used to predict the ship's position at the next moment. If the predicted position meets the preset termination condition, the path tracking is completed; otherwise, the above process is repeated.

[0069] Among them, the LOS model, PID controller and ship maneuvering motion model are existing mature technologies, and their principles will not be elaborated here.

[0070] S902. Predict the position of the target ship at a future time and calculate the minimum distance between the tracking path and the target ship.

[0071] Assuming the target ship maintains its current course and speed, its position at future time t can be predicted using the following equation:

[0072] In the formula, ( x t , y t () represents the target ship's position at the previous moment; x t-1 , y t-1 () represents the position of the target ship at a future time t; v x , v y ) represents the speed components of the target ship along the x and y axes; θ The target ship's course.

[0073] The minimum distance between the tracking path and the target ship is:

[0074] In the formula, and These represent the trajectory points of the current vessel and the target vessel at time t, respectively. t o and t f These represent the times at the start and end of the tracking path, respectively.

[0075] S903. The candidate avoidance path with the minimum distance greater than the preset safety distance and the smallest average absolute error is selected as the optimal avoidance path.

[0076] Specifically, when the minimum distance is greater than the preset safety distance, it indicates that the tracking path will not collide with the target ship, and it is safe. The smaller the mean absolute error, the more the generated avoidance path conforms to the ship's maneuverability.

[0077] The embodiments of the present invention use the candidate avoidance path with the minimum distance greater than the preset safety distance and the minimum average absolute error as the optimal avoidance path, which can ensure maneuverability and safety during the process.

[0078] In summary, the embodiments of this invention creatively combine data-driven artificial intelligence with expertise and physical models in the field of shipping to generate a ship collision avoidance decision-making method that is both intelligent and reliable, safe and feasible. This fundamentally solves the shortcomings of traditional methods in terms of practicality, interpretability and reliability, and provides key technical support for the development of intelligent shipping.

[0079] On the other hand, embodiments of the present invention also provide a data-driven ship avoidance path generation device, such as... Figure 10 As shown, the data-driven ship avoidance path generation device 1000 includes: Database construction unit 1001 is used to extract historical encounter scenarios of ships containing avoidance behavior from historical AIS data, identify historical scene features and historical avoidance features of historical encounter scenarios of ships, and construct a database based on historical scene features and historical avoidance features. The scene matching unit 1002 is used to acquire real-time scene features of the real-time navigation scene, perform similarity matching between the real-time scene features and historical scene features in the database, determine at least one target historical scene, and determine the target historical avoidance features corresponding to the target historical scene. The candidate avoidance path generation unit 1003 is used to generate at least one candidate avoidance path for a real-time navigation scenario based on historical avoidance features. The obstacle avoidance path optimization unit 1004 is used to perform maneuverability and safety assessments on at least one candidate obstacle avoidance path and determine the optimal obstacle avoidance path.

[0080] The data-driven ship avoidance path generation device 1000 provided in the above embodiments can realize the technical solutions described in the above data-driven ship avoidance path generation method embodiments. The specific implementation principles of each module or unit can be found in the corresponding content in the above data-driven ship avoidance path generation method embodiments, and will not be repeated here.

[0081] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.), and the computer program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0082] The above provides a detailed description of a data-driven ship avoidance path generation method and apparatus provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A data-driven method for generating ship collision avoidance paths, characterized in that, include: Extract historical encounter scenarios of ships containing avoidance behavior from historical AIS data, identify historical scene features and historical avoidance features of the historical encounter scenarios, and construct a database based on the historical scene features and the historical avoidance features; The real-time scene features of the real-time navigation scene are obtained, and the real-time scene features are matched with the historical scene features in the database to determine at least one target historical scene and to determine the target historical avoidance features corresponding to the target historical scene. Based on the historical avoidance features, at least one candidate avoidance path is generated for the real-time navigation scenario; Maneuverability and safety are evaluated for the at least one candidate avoidance path to determine the optimal avoidance path.

2. The data-driven ship avoidance path generation method according to claim 1, characterized in that, The historical scenario features include the encounter situation between the vessel performing the avoidance maneuver and the target vessel being avoided, as well as other features, including the nearest encounter distance, the nearest encounter time, the relative distance, the relative speed, and the relative bearing.

3. The data-driven ship avoidance path generation method according to claim 2, characterized in that, Identifying the encounter situations in the historical encounter scenarios of the vessels, including: Determine the encounter situation of the target vessel based on its relative bearing to the target vessel. Determine the encounter situation of the target ship based on the relative bearing of the ship to the target ship; The encounter situation is obtained by combining the encounter situation of the current vessel and the encounter situation of the target vessel.

4. The data-driven ship avoidance path generation method according to claim 3, characterized in that, The encounter situation described for this vessel includes categories A, B, C, D, E, F, and G. In category A, the relative bearing of the target vessel to this vessel is […]. The relative bearing of the target vessel (Category B) to this vessel is [ The relative bearing of the target vessel (Category C) to this vessel is [ The relative bearing of the target vessel (Category D) to this vessel is [ The relative bearing of the target vessel of category E to this vessel is [ The relative bearing of the target vessel of category F with respect to this vessel is [ The relative bearing of the target vessel (class G) to this vessel is [ ).

5. The data-driven ship avoidance path generation method according to claim 1, characterized in that, The historical avoidance characteristics include the starting point of the ship's avoidance behavior, the ending point of the avoidance behavior, the point where the turning rate is at its extreme value, and the trajectory point when the turning rate is zero.

6. The data-driven ship avoidance path generation method according to claim 5, characterized in that, The historical avoidance feature takes the starting point of the avoidance behavior as the extreme point, and uses radial distance and polar angle to characterize the ending point of the avoidance behavior, the trajectory point when the steering rate is at its extreme value, and the trajectory point when the steering rate is zero.

7. The data-driven ship avoidance path generation method according to claim 2, characterized in that, The real-time scene features include real-time encounter situations and other real-time features; The step of performing similarity matching between the real-time scene features and the historical scene features in the database to determine at least one target historical scene includes: Based on the real-time encounter situation, at least one candidate historical encounter scenario with the same encounter situation as the real-time encounter situation is selected from the database. The at least one target historical scenario is determined from the at least one candidate historical encounter scenario based on other features of each candidate historical encounter scenario and the other real-time features.

8. The data-driven ship avoidance path generation method according to claim 7, characterized in that, The determination of the at least one target historical scenario among the at least one candidate historical encounter scenario based on other features of each of the candidate historical encounter scenarios and the other real-time features includes: Calculate the first cumulative distribution function value of the other real-time features and the second cumulative distribution function value of the other features corresponding to each of the target historical scenes; Determine whether the absolute difference between the first cumulative distribution function value and the second cumulative distribution function value is less than a preset difference; If the value is less than the target historical scenario, then the candidate historical encounter scenario will be used as the target historical scenario.

9. The data-driven ship avoidance path generation method according to claim 1, characterized in that, The step of performing a maneuverability and safety assessment on the at least one candidate avoidance path to determine the optimal avoidance path includes: The candidate avoidance path is tracked and simulated using a ship maneuvering motion model, a line-of-sight guidance model, and a PID controller to obtain the tracking path and determine the mean absolute error between the tracking path and the candidate avoidance path. Predict the target ship's position at a future time and calculate the minimum distance between the tracking path and the target ship; The candidate avoidance path that has the minimum distance greater than the preset safety distance and the minimum average absolute error is selected as the optimal avoidance path.

10. A data-driven ship avoidance path generation device, characterized in that, include: The database construction unit is used to extract historical encounter scenarios of ships containing avoidance behavior from historical AIS data, identify historical scene features and historical avoidance features of the historical encounter scenarios, and construct a database based on the historical scene features and the historical avoidance features. The scene matching unit is used to acquire real-time scene features of a real-time navigation scene, perform similarity matching between the real-time scene features and historical scene features in the database, determine at least one target historical scene, and determine the target historical avoidance features corresponding to the target historical scene. A candidate avoidance path generation unit is used to generate at least one candidate avoidance path for the real-time navigation scenario based on the historical avoidance features. The obstacle avoidance path optimization unit is used to perform maneuverability and safety assessments on the at least one candidate obstacle avoidance path and determine the optimal obstacle avoidance path.

Images