Ship collision warning method, device and equipment

Abstract

The application discloses a ship collision early warning method, device and equipment, and relates to the field of early warning, which comprises the following steps: acquiring dynamic and static data and target detection data of each ship in a channel, performing space-time registration and fusion, generating an optimal estimated trajectory for each ship, performing short-term trajectory prediction according to the optimal estimated trajectory, and outputting predicted positions and speeds at multiple future moments; inputting the dynamic and static data into a ship dynamics model to obtain a theoretical braking distance and a minimum turning radius; calculating a dynamic safety boundary for each ship based on the optimal estimated trajectory, the theoretical braking distance and the minimum turning radius; performing collision risk prediction on the optimal estimated trajectory, the predicted positions, the predicted speeds and the dynamic safety boundary through a preset rule and a machine learning model, and identifying a collision risk result between ships; and generating safety recommended speeds and hierarchical early warning information for a target ship according to the collision risk result and sending the information to a corresponding ship terminal. The application improves the collision early warning accuracy.

Description

Technical Field

[0001] This application relates to the field of early warning technology, and in particular to a method, device and equipment for early warning of ship collisions. Background Technology

[0002] With the rapid development of inland waterway shipping, the number of vessels has continued to increase, vessel tonnage has gradually become larger, and navigation density has significantly improved. Coupled with the complex and variable environment of inland waterways, such as numerous winding channels, abundant shoals and reefs, frequent bridge and tunnel crossings, unstable water flow velocity and direction, and significant seasonal water level fluctuations, the risk of collisions between inland waterway vessels has increased dramatically. Therefore, in order to reduce the incidence of collision accidents and ensure shipping safety, research on identifying inland waterway vessel collisions and conducting risk warnings is crucial.

[0003] Currently, relevant technologies rely on Automatic Identification System (AIS) and radar data to calculate the Distance of Closest Point of Approach (DCPA) and Time to Closest Point of Approach (TCPA) between two ships. A collision warning is triggered when these two parameters fall below corresponding preset thresholds. However, this approach only depends on fixed thresholds and does not consider the differences in braking performance and collision avoidance requirements between ships of different tonnages, such as 10,000-ton cargo ships and 1,000-ton barges, resulting in low accuracy of collision warnings. Summary of the Invention

[0004] The purpose of this application is to provide a ship collision warning method, device, and equipment to solve the technical problem that the collision warning in the prior art is too general and one-sided, resulting in low accuracy of collision warning.

[0005] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for ship collision early warning, including: Acquire dynamic and static data and target detection data of each vessel in the waterway; The dynamic and static data and the target detection data are spatiotemporally registered and fused to generate an optimal estimated trajectory for each ship. Based on the optimal estimated trajectory, a Kalman filter is used to predict the short-term trajectory and output the predicted position and predicted speed at multiple future times. The dynamic and static data are input into the trained ship dynamics model to obtain the theoretical braking distance and minimum turning radius of the ship under the current load and waterway conditions. Based on the optimal estimated trajectory, the theoretical braking distance, and the minimum turning radius, a dynamic safety boundary is calculated for each vessel; the dynamic safety boundary is an asymmetric elliptical region centered on the vessel's hull that changes over time. The optimal estimated trajectory, predicted position, predicted speed, and dynamic safety boundary are used to predict collision risk and identify collision risk results between ships through preset rules and machine learning models. Based on the collision risk results, a safe recommended speed and graded early warning information are generated for the target vessel and sent to the corresponding vessel terminal; the target vessel is a vessel with a collision risk.

[0006] Secondly, this application provides a ship collision warning device, which includes: The acquisition module is used to acquire dynamic and static data and target detection data of each ship in the waterway; The processing module is used to perform spatiotemporal registration and fusion of the dynamic and static data and the target detection data, generate the optimal estimated trajectory for each ship, and use a Kalman filter to perform short-term trajectory prediction based on the optimal estimated trajectory, and output the predicted position and predicted speed at multiple future times. The first determining module is used to input the dynamic and static data into the trained ship dynamics model to obtain the theoretical braking distance and minimum turning radius of the ship under the current load and waterway conditions. The second determining module is used to calculate the dynamic safety boundary for each ship based on the optimal estimated trajectory, the theoretical braking distance, and the minimum turning radius; the dynamic safety boundary is an asymmetric elliptical region centered on the ship's hull that changes over time. The risk identification module is used to predict collision risks between ships by using the optimal estimated trajectory, predicted position, predicted speed and dynamic safety boundary through preset rules and machine learning models; The collision warning module is used to generate a safe recommended speed and graded warning information for the target vessel based on the collision risk results and send it to the vessel terminal corresponding to the target vessel; the target vessel is a vessel with a collision risk.

[0007] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the ship collision warning method described in any one of the above.

[0008] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a ship collision early warning method, device, and equipment. Compared with existing technologies, this solution achieves short-term trajectory prediction by fusing the ship's dynamic and static data with target detection data and using Kalman filtering. This allows for a more comprehensive capture of the ship's real-time and future motion states, providing data guidance for subsequent risk assessment and avoiding misjudgments caused by single sensor data bias. Furthermore, by utilizing a trained ship dynamics model and combining the ship's own dynamic and static data, it outputs personalized theoretical braking distances and minimum turning radii. This overcomes the limitations of traditional technologies that rely solely on fixed thresholds, fully adapting to the differences in braking performance and collision avoidance requirements of ships of different tonnages, such as 10,000-ton cargo ships and 1,000-ton barges, fundamentally improving the adaptability of risk assessment. Based on optimal trajectory estimation and theoretical... By using braking distance and minimum turning radius, an asymmetric elliptical dynamic safety boundary centered on the hull is constructed, accurately depicting the safety risk area that changes with the ship's motion state. Compared to traditional fixed geometric boundaries, this more closely reflects the risk distribution patterns in actual navigation. Furthermore, through the collaborative analysis of preset rules and machine learning models, the false alarm and false alarm rates are effectively reduced, greatly improving the accuracy of risk identification while adapting to complex inland waterway encounter scenarios. Finally, based on the risk results, recommended safe speeds and graded early warning information are generated, transforming abstract risk judgments into specific, actionable operational guidelines sent to the ship's terminal. This facilitates rapid and accurate collision avoidance decisions by the crew, significantly improving the timeliness and scientific rigor of collision avoidance operations, enhancing the accuracy of collision warnings, and ensuring the safe and efficient operation of inland waterway transportation. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A schematic diagram of a ship collision warning system provided in one embodiment of this application; Figure 2 This is a schematic diagram of the structure of a ship collision warning system provided in another embodiment of this application; Figure 3 A schematic flowchart of a ship collision warning method provided in an embodiment of this application; Figure 4 A flowchart illustrating a method for determining the theoretical braking distance and minimum turning radius, provided in another embodiment of this application; Figure 5 A flowchart illustrating the determination of theoretical braking distance and minimum turning radius provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a dynamic security boundary provided in an embodiment of this application; Figure 7 A schematic flowchart illustrating a method for identifying collision risk results provided in an embodiment of this application; Figure 8 This is a schematic diagram of the functional modules of a ship collision warning device provided in an embodiment of this application; Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] For ease of understanding, some technical terms involved in the embodiments of this application are explained below: An Automatic Identification System (AIS) is a maritime navigation and communication device installed on ships. Utilizing the Global Positioning System (GPS) and communication technology, it enables ships to automatically transmit information such as their position, course, and speed, and exchange this information in real time with surrounding vessels and shore-based stations. Used for collision avoidance, it automatically and continuously broadcasts static, dynamic, voyage, and safety-related information to surrounding vessels and shore-based stations via the Very High Frequency (VHF) radio band. It is an important navigational aid and monitoring tool in the maritime field.

[0014] The Closest Point of Approach (CPA) is the point at which another vessel is closest to the ship as it passes. During the relative motion of the radar, this point is determined by the foot of the perpendicular line between the center of the radar chart and the relative motion line of the other vessel. From this, the Distance of Closest Point of Approach (DCPA) and the Time to Closest Point of Approach (TCPA) can be calculated. In modern ship collision avoidance systems, radars equipped with ARPA (Automatic Radar Plotting) acquire DCPA and TCPA information by tracking the target vessel's motion and fuse this information with AIS (Automatic Information System) perception information to improve calculation accuracy. Collision avoidance algorithms based on COLREGs rules generate adaptive collision avoidance trajectories by integrating DCPA, TCPA, and other factors, and their effectiveness is verified through ship simulation experiments.

[0015] Meeting in opposite directions: refers to the meeting of two ships in the course of a course, including meeting in opposite directions or close to meeting, meeting from the port or starboard side, meeting in a curved course, but excluding the meeting of two ships crossing over.

[0016] Related technologies employ ship collision warning systems based on the fusion of AIS and radar data. This system tracks ship targets by integrating information such as ship identity, heading, and speed acquired from AIS with target position data detected by radar. It then calculates the Distance to Encounter (DCPA) and Time to Encounter (TCPA), triggering a collision alarm when these two parameters fall below preset fixed thresholds. However, this approach has significant drawbacks: first, it uses a one-size-fits-all fixed safety threshold, failing to consider the differences in braking performance and collision avoidance requirements between ships of different tonnages, such as 10,000-ton cargo ships and 1,000-ton barges, resulting in insufficient warning accuracy; second, the warning information is only a general collision warning. The collision warning system lacks personalized collision avoidance suggestions for specific encounter scenarios, failing to effectively assist vessel operators in making reasonable collision avoidance decisions. Furthermore, it has poor adaptability to typical inland waterway encounter situations such as complex intersections, failing to optimize collision risk assessment logic based on unique inland waterway factors such as channel boundary constraints and water flow influences, leading to false alarms or missed alarms. Additionally, the system relies on the fusion of AIS and radar data, still unable to address issues such as blind spots caused by missing or abnormal AIS equipment on small inland waterway vessels, and warning failures due to decreased sensor data accuracy under adverse weather conditions. Overall, it is ill-suited to the safety warning needs of complex inland waterway shipping scenarios.

[0017] To address the aforementioned shortcomings, this application provides a ship collision early warning method. Compared to existing technologies, this solution integrates the ship's dynamic and static data with target detection data and uses Kalman filtering to achieve short-term trajectory prediction. This allows for a more comprehensive capture of the ship's real-time and future motion states, providing data guidance for subsequent risk assessment and avoiding misjudgments caused by single-sensor data bias. Furthermore, by leveraging a trained ship dynamics model and combining the ship's own dynamic and static data, it outputs personalized theoretical braking distances and minimum turning radii. This overcomes the limitations of traditional technologies that rely solely on fixed thresholds, fully adapting to the differences in braking performance and collision avoidance requirements of ships of different tonnages, such as 10,000-ton cargo ships and 1,000-ton barges, thereby fundamentally improving the adaptability of risk assessment. Finally, based on the optimal estimated trajectory, theoretical braking distance, and minimum turning radius... This system constructs an asymmetric elliptical dynamic safety boundary centered on the ship's hull, accurately depicting the safety risk areas that change with the ship's motion state. Compared to traditional fixed geometric boundaries, this system better reflects the risk distribution patterns in actual navigation. Furthermore, through the collaborative analysis of preset rules and machine learning models, it effectively reduces false alarms and missed alarms, significantly improving the accuracy of risk identification while adapting to complex inland waterway encounter scenarios. Finally, based on the risk results, it generates recommended safe speeds and graded early warning information, transforming abstract risk judgments into concrete and actionable operational guidelines sent to the ship's terminal. This facilitates rapid and accurate collision avoidance decisions by the crew, significantly improving the timeliness and scientific rigor of collision avoidance operations, enhancing the accuracy of collision warnings, and ensuring the safe and efficient operation of inland waterway transportation.

[0018] The ship collision warning method provided in this application embodiment can be applied to, for example... Figure 1 The application environment of the ship collision warning method shown is as follows. This application environment includes a ship collision warning system, which comprises: a collision warning device 10 and at least one ship terminal 20. The collision warning device 10 establishes a communication connection with each ship terminal 20. The at least one ship terminal may include a first ship terminal, a second ship terminal, ..., an nth ship terminal, where n is a positive integer.

[0019] The collision warning device 10 is used to acquire dynamic and static data and target detection data of each vessel in the waterway, generate a safe recommended speed and graded warning information for the target vessel, and send it to the corresponding vessel terminal. The vessel terminal 20 is used to receive the safe recommended speed and graded warning information, navigate at the safe recommended speed, and issue collision warnings based on the graded warning information.

[0020] Optionally, please see Figure 2As shown, the collision warning device 10 is equipped with a shore-based sensing module 11 and a shore-based processing center 12. The shore-based sensing module 11 may include various data acquisition devices, such as millimeter-wave radar, AIS base station network and camera. The shore-based processing center 12 is used to generate safe recommended speed and graded warning information for the target vessel based on the dynamic and static data and target detection data of each vessel in the waterway and send it to the corresponding vessel terminal of the target vessel.

[0021] Among them, the shore-based perception module 11 is used to collect multi-source traffic data of various ships in the waterway; the shore-based processing center 12 includes: a data fusion and trajectory tracking module, a ship dynamics model library, a dynamic safety boundary calculation module, a collision risk assessment engine, and a speed guidance and early warning generation module.

[0022] The data fusion and trajectory tracking module performs spatiotemporal registration and fusion of dynamic and static data and target detection data to generate an optimal estimated trajectory for each ship. Based on the optimal estimated trajectory, it uses a Kalman filter to perform short-term trajectory prediction, outputting the predicted position and speed at multiple future moments. The ship dynamics model library provides ship dynamics models for determining the theoretical braking distance and minimum turning radius of the ship. The dynamic safety boundary calculation module calculates the dynamic safety boundary for each ship based on the optimal estimated trajectory, theoretical braking distance, and minimum turning radius. The collision risk assessment engine predicts collision risks using preset rules and machine learning models, identifying collision risk outcomes between ships. The speed guidance and warning generation module generates recommended safe speeds and graded warning information for target ships based on collision risk outcomes.

[0023] The aforementioned ship terminal 20 is installed on the ship and establishes a communication connection with the shore-based sensing module 11. It receives recommended safe speed and graded early warning information to navigate at the recommended safe speed and to issue collision warnings based on the graded warning information. The ship terminal includes an onboard microcomputing unit and sensors: the onboard microcomputing unit can independently or assist in running a simplified version of the dynamic safety boundary calculation module and risk assessment engine, providing the ship with autonomous early warning capabilities based on local perception when shore-based communication is interrupted. It also features a human-machine interface for human-machine interaction.

[0024] In one exemplary embodiment, such as Figure 3 As shown, a ship collision warning method is provided. This method is executed by a collision warning device, and can be executed by a computer device such as a terminal or server alone, or by both a terminal and a server. In this embodiment, it includes the following steps S201 to S206. Wherein: Step S201: Obtain dynamic and static data and target detection data of each ship in the waterway.

[0025] Specifically, the aforementioned dynamic and static data refers to various types of data acquired through multiple data acquisition devices that reflect the ship's operational status. These data acquisition devices include radar, AIS base stations, and draft sensors. The data acquired by the AIS base station is AIS data, and the data acquired by the radar is target detection data, which includes the ship's position (…). ), velocity vector This dynamic and static data includes: vessel identification number (MMSI), vessel type. Ship tonnage Structural property parameters, real-time speed, channel conditions, and real-time draft; real-time draft. It is obtained through a draft sensor. Structural property parameters refer to the ship's structural attributes, including length. , ship width Etc. Static and dynamic data can also include dynamic real-time ground speed. Real-time heading to the ground .

[0026] The radar used can be millimeter-wave radar, which detects ships in the waterway, obtains corresponding point cloud data, and analyzes the point cloud data to obtain target detection data, which may include the real-time position coordinates of the ships. Real-time draft is a core parameter reflecting the actual load state of the ship, directly affecting its maneuverability, such as braking and turning.

[0027] This step combines dynamic and static data with target detection data, enabling a comprehensive consideration of data information from different devices and providing high-quality and reliable data for subsequent risk assessment.

[0028] Step S202 involves performing spatiotemporal registration and fusion of dynamic and static data and target detection data to generate an optimal estimated trajectory for each ship. Based on the optimal estimated trajectory, a Kalman filter is used to perform short-term trajectory prediction, outputting the predicted position and predicted speed at multiple future times.

[0029] It is understandable that dynamic and static data and target detection data have inherent differences in acquisition principles, update frequencies, and coordinate references. For example, AIS data update cycles are usually a few seconds to tens of seconds, while radar data can be updated to the millisecond level. Furthermore, the two may use different coordinate systems, such as AIS data commonly using the WGS-84 geographic coordinate system, while radar target detection data commonly using a relative coordinate system. Direct fusion would result in data redundancy and contradictions. Therefore, it is necessary to perform spatiotemporal registration and fusion of dynamic and static data and target detection data.

[0030] Specifically, after acquiring dynamic and static data and target detection data, it is necessary to unify the spatiotemporal reference of the data. In the time dimension, interpolation or resampling algorithms are used to align data with different update frequencies to a unified time axis. In the spatial dimension, coordinate transformation algorithms are used to map the radar's relative coordinates to the AIS geographic coordinate system, ensuring that both types of data describe the ship's state in the same spatiotemporal context. After completing spatiotemporal registration, data fusion algorithms are used to integrate the AIS data and target detection data. High-frequency position data from the target detection data ensures the real-time performance and continuity of the trajectory, while speed and heading data from the AIS data constrain the trajectory's motion trend. Ultimately, a smooth and accurate optimal estimated trajectory is generated for each ship. Data fusion algorithms include, for example, weighted fusion and Bayesian estimation algorithms.

[0031] Furthermore, after determining the optimal estimated trajectory, a state transition model adapted to scenarios of uniform straight-line movement and lane-changing is constructed by combining the ship's navigation characteristics and defining core motion state variables including the ship's geographic coordinates and directional velocity components. Simultaneously, the observation model is matched to clarify the correspondence between the position data of the optimal estimated trajectory and the ship's state variables. Then, filter initialization is performed. Position data from consecutive historical moments are extracted from the optimal estimated trajectory, and the initial velocity is calculated by the ratio of adjacent position changes to the time interval. The initial position and initial velocity are used as the initial state of the Kalman filter. The initial state covariance matrix is ​​set based on the sensor measurement accuracy; this covariance matrix characterizes the uncertainty of the initial state. The filter's initial state (initial position, initial velocity) and the initial state covariance matrix serve as the starting point reference data for Kalman filter prediction and update iteration. The process involves performing an update and iterative optimization operation. Based on the optimal state and state transition model from the previous time step, the predicted state and corresponding covariance for the current time step are calculated. The position data of the optimal estimated trajectory at the current time step are used as observations, and the deviation between this deviation and the predicted state is calculated. The Kalman gain is then calculated using the covariance matrix, and this Kalman gain is used to correct the predicted state, resulting in the optimal posterior state for the current time step. This iteration is repeated until all historical data of the optimal estimated trajectory is exhausted, thereby obtaining the optimal state of the ship at the current time step and the posterior covariance matrix. The optimal state of the ship includes the predicted position and predicted speed. The prediction time step and total duration can be preset. Based on the optimal state at the current time step and the state transition model, multi-step iterative extrapolation is performed to sequentially calculate the ship's state at each future time step. The predicted position and predicted speed for multiple future time steps are extracted from the states at each future time step.

[0032] In this step, by performing spatiotemporal registration of dynamic and static data and target detection data, data from different sensors and different time dimensions can be unified into the same spatiotemporal coordinate system, eliminating data redundancy and contradictions. Then, the optimal estimation algorithm of the Kalman filter is used to process the fused data. Through the iterative process of prediction and update, sensor noise in harsh environments is filtered out, and the predicted position and speed of the ship at several future moments are output. This makes the generated ship trajectory smoother and more accurate, providing a reliable trajectory basis for subsequent collision risk assessment.

[0033] Step S203: Input the dynamic and static data into the trained ship dynamics model to obtain the theoretical braking distance and minimum turning radius of the ship under the current load and waterway conditions.

[0034] It should be noted that the aforementioned ship dynamics model is a pre-trained ship tonnage-speed-braking distance model. This mathematical model, based on ship static parameters, real-time navigation status, and environmental conditions, is used to accurately calculate key maneuvering performance indicators such as the ship's individualized theoretical braking distance and minimum turning radius. It includes a braking distance model and a minimum turning radius model. The braking distance model is used to calculate the ship's theoretical braking distance. The minimum turning radius model is used to calculate the minimum turning radius of a ship. The theoretical braking distance is the distance a ship travels from the issuance of a braking command to a complete stop. The minimum turning radius is the minimum turning radius required for a ship to maintain stable navigation. These two parameters directly determine the collision avoidance limits of a ship. For example, the braking distance of a 10,000-ton cargo ship is much greater than that of a 1,000-ton barge. If the same safety threshold is used, the cargo ship will be unable to avoid a collision in time due to insufficient braking distance, while the barge will generate false warnings due to an overly strict threshold.

[0035] By calculating the theoretical braking distance and minimum turning radius of a ship using an individualized ship dynamics model, the limitations of relying solely on threshold values ​​are avoided. This approach allows for the calculation of dynamic parameters tailored to the specific physical characteristics of each vessel, overcoming the limitation of existing technologies that use fixed thresholds and cannot adapt to ships of different tonnages. The ship dynamics model can be constructed by collecting a large amount of actual braking distance and maneuverability data for ships at different tonnages, drafts, and speeds. Multiple nonlinear regression or machine learning algorithms are then used, with tonnage, draft, initial speed, and meteorological data samples as core input features, and braking distance and turning performance indicators as output targets, to train the model. This ship dynamics model provides personalized dynamic parameter estimates that conform to the physical characteristics of different ships.

[0036] The aforementioned ship dynamics model can be constructed through the following steps: The system acquires multi-source operational data of the vessel and preprocesses the data to obtain processed data. Preprocessing includes data cleaning and data labeling. Processed data includes vessel static parameters, meteorological and hydrological data samples, actual braking distance and actual minimum turning radius. Vessel static parameters include vessel tonnage samples, draft samples, initial speed samples, channel condition samples, and structural attribute samples.

[0037] Meteorological and hydrological data samples, ship tonnage samples, draft samples, initial speed samples, and channel condition samples are input into the initial braking distance model to obtain the predicted braking distance. Meteorological and hydrological data samples, initial speed samples, and structural attribute samples are input into the initial minimum turning radius model to obtain the predicted minimum turning radius.

[0038] A loss function is constructed based on the predicted braking distance and the actual braking distance, as well as based on the predicted minimum turning radius and the actual minimum turning radius.

[0039] By minimizing the loss function, machine learning algorithms are used to train the model parameters of the initial braking distance model and the initial minimum turning radius model to obtain the ship dynamics model.

[0040] Specifically, the aforementioned multi-source operational data can include different data sources to ensure data coverage of different ship types, loads, and navigation scenarios. These sources may include historical AIS data, test data, and data collected from pilot vessels. For historical AIS data, feature data such as speed abrupt changes and trajectory curvature changes can be extracted from the ship's historical AIS database. Inverse dynamics algorithms can then be used to infer the ship's braking behavior and turning maneuvers, compensating for the high cost and insufficient scenario coverage of actual ship testing. For test data, braking and turning tests under typical conditions such as calm water, upstream currents, and crosswinds can be simulated in a standardized simulator environment for ships of different tonnages and drafts, such as thousand-ton barges and ten-thousand-ton cargo ships, to obtain accurate maneuverability benchmark data. For data collected from pilot vessels, shipboard data recorders can be installed on the pilot vessels to collect real-time maneuvering data such as ship speed, steering angle, and main engine power changes during actual navigation.

[0041] After acquiring multi-source operational data, it undergoes preprocessing, including data cleaning and labeling. During data cleaning, abnormal data points caused by sensor malfunctions or signal loss are removed, such as sudden speed changes exceeding the ship's physical performance limits or discontinuities in trajectory data, ensuring the authenticity of the input data. During data labeling, standardized labels are added to each braking or turning data record. Core labeling dimensions include ship static parameters, meteorological and hydrological data samples, actual braking distance, and actual minimum turning radius, providing clear input-output mapping samples for model training. Ship static parameters include ship tonnage samples, draft samples, and initial speed samples. Waterway condition samples, structural attribute samples, etc.

[0042] To address the diverse needs of inland waterway transportation, a model is trained using multiple nonlinear regression and machine learning algorithms, balancing interpretability and high accuracy. First, a multiple nonlinear regression algorithm is used to construct the initial braking distance model and the initial minimum turning radius model. The initial braking distance model can be expressed by the following formula: ; The initial minimum turning radius model is expressed by the following formula: ; in, The theoretical braking distance, For draft depth samples, For the initial speed sample, For waterway condition samples, For the width of the boat, These are the model parameters determined through regression analysis.

[0043] Optionally, in addition to the ship's inherent parameters and real-time state parameters, the input variables of the above model can also include meteorological and hydrological data samples, and environmental feature vectors can be used to... express, ,in, For wind speed, For wind direction, For water flow velocity, In terms of water flow direction, For visibility. By introducing environmental features, the model can calculate individualized theoretical braking distances and minimum turning radii under different wind, current, and visibility conditions. By introducing environmental feature vectors, the dynamic model can be extended from still water and standard environments to complex and variable external conditions, which significantly improves the model's practicality and accuracy. This enables the safety model to dynamically respond to changes in meteorology and hydrology, which is the core prerequisite for achieving high-precision safety early warning.

[0044] Gradient Boosting Decision Tree (GBDT) or Deep Neural Network (DNN) algorithms are used to achieve accurate fitting of complex nonlinear relationships, obtaining the output values ​​of the initial braking distance model and the initial minimum turning radius model, which are the predicted braking distance and the predicted minimum turning radius, respectively. Based on the predicted braking distance and the actual braking distance, and based on the predicted minimum turning radius and the actual minimum turning radius, the mean square error is determined to construct the loss function. The model is trained according to minimizing the loss function to optimize the model parameters, thus obtaining the braking distance model and the minimum turning radius model.

[0045] In this step, by pre-training the ship dynamics model, it is possible to dynamically adapt to the real-time state of each ship, making the output dynamic parameters more personalized and accurate.

[0046] In one embodiment, a specific implementation method for determining the theoretical braking distance and minimum turning radius of a vessel under current load and channel conditions is also provided; please refer to [link to relevant documentation]. Figure 4 As shown, the method includes: Step S301: Obtain meteorological and hydrological data of the current environment of the ship.

[0047] Step S302: Input meteorological and hydrological data, ship tonnage, real-time draft, real-time speed and channel conditions as feature vectors into the braking distance model to output the theoretical braking distance. Input meteorological and hydrological data, real-time speed and structural attribute parameters into the minimum turning radius model to obtain the minimum turning radius of the ship under the current load and channel conditions.

[0048] In this embodiment, the meteorological and hydrological data can be imported from external devices, obtained from blockchain or databases, or obtained through real-time detection. This embodiment does not limit the method of obtaining meteorological and hydrological data of the ship's current environment.

[0049] Please see Figure 5 As shown, when there are multiple vessels in the waterway, after acquiring the meteorological and hydrological data, vessel tonnage, real-time draft, real-time speed, and waterway conditions for each vessel, these data are used as feature vectors and input into the deployed ship dynamics model to obtain the theoretical braking distance of each vessel under the current condition. and minimum turning radius .

[0050] It is understandable that the above-mentioned ship dynamics model has an online update mechanism: 1) the system continuously collects new ship maneuvering data; 2) when the new data accumulates to a certain scale or the model prediction error continues to exceed the threshold, the incremental learning or retraining process of the model is triggered so that the model parameters can adapt to the evolution of the overall ship maneuvering performance in the waterway. The online update mechanism gives the system the ability to learn and evolve on its own, and can adapt to the impact of ship upgrades or changes in crew operating habits.

[0051] This embodiment, by comprehensively considering multiple influencing factors such as meteorological and hydrological data, ship parameters, real-time navigation status, and waterway conditions, breaks through the limitations of traditional ship braking distance and minimum turning radius calculations that rely on fixed empirical values. It can accurately output individualized maneuvering performance indicators adapted to the ship's current actual operating conditions. It not only fully considers the differences in maneuvering characteristics of ships of different tonnages and loads, but also takes into account the impact of environmental factors such as wind and current on the ship's motion state. It provides core parameter support that fits the actual navigation scenario for the subsequent construction of dynamic safety boundaries and accurate assessment of collision risks, effectively improving the adaptability and reliability of the inland waterway ship collision early warning system.

[0052] Step S204: Based on the optimal estimated trajectory, theoretical braking distance and minimum turning radius, calculate the dynamic safety boundary for each ship; the dynamic safety boundary is an asymmetric elliptical region centered on the ship's hull that changes over time.

[0053] The aforementioned Dynamic Safety Boundary (DSB) is an asymmetrical elliptical region, including its major and minor axes. The major axis of the ellipse aligns with the ship's bow. Its core function is to cover the safety distance required for longitudinal braking. Its length calculation is directly related to the ship's braking performance and real-time speed, preventing collisions caused by delayed braking. The minor axis of the ellipse is perpendicular to the major axis, i.e., the ship's lateral direction. Its core function is to cover the lateral safety space during turning and obstacle avoidance. Its basic length calculation is related to the ship's turning performance and real-time turning status. By upgrading the traditional fixed safety distance to an asymmetrical elliptical region that changes in real-time with the ship's motion (speed, turning rate), it better reflects the actual maneuvering characteristics of the ship.

[0054] In one embodiment, a specific implementation method for calculating the dynamic safety boundary for each vessel based on the optimal estimated trajectory, theoretical braking distance, and minimum turning radius is also provided, including: Based on the optimal estimated trajectory, the real-time turning rate of the vessel is determined; the basic length of the major axis of the elliptical region is calculated based on the real-time speed and theoretical braking distance; the basic length of the minor axis of the elliptical region is calculated based on the minimum turning radius and the real-time turning rate; the basic lengths of the major and minor axes are corrected based on meteorological and hydrological data to obtain the corrected lengths of the major and minor axes; the left and right minor axis lengths of the elliptical region are calculated based on the sign of the real-time turning rate and the corrected lengths of the minor axes; the vessel's center of mass is used as the coordinates of the center point of the elliptical region, the vessel's heading angle is used as the major axis direction angle of the elliptical region, and the corrected lengths of the major axis, the left minor axis length, and the right minor axis length are used as the major and minor axes of the ellipse to determine the dynamic safety boundary.

[0055] Specifically, after obtaining the optimal estimated trajectory, the ship's position coordinates at three consecutive adjacent moments are extracted from the optimal estimated trajectory. First, the ship's initial heading angle is calculated using the position coordinates at the first two moments, and then the ship's current heading angle is calculated using the position coordinates at the last two moments. Subsequently, the difference between the two heading angles is calculated to obtain the heading angle change. At the same time, the time interval between adjacent moments is determined. Finally, the ratio of the heading angle change to the time interval is calculated to obtain the ship's real-time turning rate. When the real-time turning rate is negative, it indicates that the ship is turning left, and when the real-time turning rate is positive, it indicates that the ship is turning right, which is used to reflect the intensity of the current turning operation.

[0056] After obtaining the real-time turning rate, real-time speed, theoretical braking distance, minimum turning radius, and real-time turning rate, the real-time speed and theoretical braking distance are calculated using the major axis length calculation function. The length of the major axis foundation can be calculated using the following formula: ; in, , The weighting coefficient is greater than 0 and can be customized according to actual needs; This serves as the baseline safety length constant. This is the theoretical braking distance; For the ship's real-time speed, This is the length of the major axis base.

[0057] The minimum turning radius and real-time steering ratio are calculated using the minor axis length function. The minor axis foundation length can be calculated using the following formula: ; in, , The weighting coefficient is greater than 0 and can be customized according to actual needs; The baseline safety width constant; Minimum turning radius; For the ship's real-time turning rate, This is the base length of the short axis.

[0058] It is understandable that environmental factors such as wind and current in inland waterways can significantly alter a vessel's trajectory and maneuverability. For example, braking distance increases when sailing against the current, and crosswinds can cause lateral drift. Traditional solutions do not consider these factors and are prone to failure in harsh environments. This solution considers meteorological and hydrological data when determining the dynamic safety boundary, which is achieved through the environmental feature vector E. env This indicates, including wind speed Water flow velocity By applying a second correction to the basic lengths of the major and minor axes using environmental feature vectors, the corrected lengths of the major and minor axes are obtained, expressed by the following formula: ; ; in, For wind speed, For water flow velocity, These are the influence coefficients determined based on the wind, current, and the relative direction of the ship. For the short axis base length, For the length of the major axis base, To correct the length of the major axis, The length is adjusted for the short axis.

[0059] This step involves incorporating environmental factors to correct the major and minor axes, upgrading the safety boundary from a static geometric shape to a dynamic adaptive boundary. This boundary can adjust in real time according to changes in wind and current intensity and direction. For example, in strong crosswinds, the minor axis length automatically increases to cover the risk of lateral drift, while in countercurrent conditions, the major axis automatically extends to accommodate longer braking distances, significantly improving the accuracy and safety of the safety boundary under complex operating conditions. This correction transforms the concept of "risk asymmetry" from a theoretical one into quantifiable and executable mathematical logic, enabling the safety boundary to accurately match the actual risk distribution during ship turning, greatly reducing false alarms and missed alarms in turning scenarios.

[0060] During a ship's turn, the collision risks on the inside and outside sides differ significantly. The inside side has a smaller turning radius and a more concentrated risk, while the outside side has a larger turning trajectory deviation and a wider risk coverage. Traditional symmetrical short axis designs cannot accurately match this scenario. This solution implements asymmetric correction based on the real-time turning rate to ensure that the boundaries perfectly match the actual risk distribution, thereby calculating the lengths of the left and right short axes.

[0061] Specifically, when the real-time steering ratio ω < 0 (left turn), the left side is the inside of the steering wheel, and the left short axis length... , The inner scaling factor reduces inner redundancy to fit the actual steering trajectory; the right side represents the outer steering direction, and the right short axis length... Maintain the basic length to cover the risk of outward deviation. When the real-time steering ratio ω > 0 (right turn), the right side is the inside of the steering wheel, and the left side is the shorter axis length. right minor axis length When the real-time turning rate ω=0 (straight ahead), the ship's lateral risk is symmetrical, therefore the length of the port minor axis is... =Right minor axis length = .

[0062] After performing secondary corrections on the major axis and minor axis base lengths using environmental feature vectors to obtain the corrected major axis and minor axis lengths, the minor axis base length in the formula can be used when calculating the left and right minor axis lengths. Replace with short axis correction length Thus, the lengths of the left and right minor axes corresponding to the modified environmental feature vector are obtained.

[0063] A ship's lateral safety requirements are directly related to its turning capability. Ships with smaller minimum turning radii have higher lateral maneuverability, allowing for a moderate reduction in the basic short axis. Conversely, a larger absolute value of the turning rate, indicating more aggressive turning maneuvers, increases the risk of lateral drift, necessitating a corresponding increase in the basic short axis. This ensures that the ship does not collide with surrounding vessels or channel boundaries during turning, laying the foundation for subsequent asymmetric corrections. (The major axis of the ellipse...) Located on the bow-stern line of the ship, its length is positively correlated with the theoretical braking distance and the current speed; the minor axis of the ellipse is perpendicular to the major axis, and its length is related to the minimum turning radius and the current turning rate, with the length of the minor axis on the inside of the turn being... (Left minor axis length) can be less than the outer minor axis length (Right short axis length) is used to illustrate the asymmetric risk during steering; the corresponding dynamic safety boundary diagram can be found as follows: Figure 6 As shown.

[0064] In this embodiment, to ensure that the dynamic safety boundary can be efficiently integrated with subsequent risk assessment algorithms, the calculation results need to be encapsulated into a standardized data structure, clearly defining the digital output format. This data structure contains several key parameters: 1) the coordinates of the ellipse's center point, which coincide with the ship's center of mass, ensuring precise binding between the boundary and the ship's real-time position; 2) the correction length of the ellipse's major axis. 3) The port and starboard minor axis lengths after asymmetric correction; 4) The direction angle of the major axis of the ellipse, consistent with the ship's heading angle, ensuring that the boundary deflects synchronously with the ship's course. By standardizing and encapsulating the key parameters of the dynamic safety boundary, the interface adaptation problem caused by inconsistent data formats is eliminated, enabling the risk assessment algorithm to directly call this data structure and quickly complete core risk judgments such as whether its own boundary overlaps with the boundaries of surrounding ships, and the size of the overlapping area, thereby improving the operational efficiency of the entire early warning system.

[0065] Optionally, the vessel's navigation status and environmental conditions are constantly changing, requiring the dynamic safety boundary to be updated synchronously to ensure timely risk coverage. The update mechanism is linked to the periodic data acquisition, trajectory tracking, and dynamic parameter calculation processes. Whenever new dynamic or static data is acquired, new speed / turning rate is output, or individualized dynamic parameters are updated, the entire calculation process in this step is immediately triggered, generating a new dynamic safety boundary and updating the data structure. This real-time update mechanism ensures that the safety boundary remains consistent with the vessel's actual navigation status and surrounding environment, avoiding misjudgments of risk due to boundary lag and providing a guarantee for continuous and accurate collision warnings.

[0066] Step S205 involves using the optimal estimated trajectory, predicted position, predicted speed, and dynamic safety boundary to predict collision risk and identify collision risk results between ships through preset rules and machine learning models.

[0067] It should be noted that the aforementioned preset rules can be customized according to actual needs. They can be based on the International Code for Preventing Collisions at Sea (COLREGs), calculating the closest encounter distance (DCPA) and time to closest encounter point (TCPA) between two vessels and comparing them with preset thresholds for different encounter scenarios to determine the existence of collision risk. The machine learning model is a pre-trained classifier. Vectors containing features such as relative heading, speed, vessel type combination, and dynamic safety boundary overlap area are input into the machine learning model, which outputs a collision probability value. These preset rules can be deployed on a rule engine. By utilizing the rule engine and machine learning model in collaborative processing, it is possible to determine in real time whether there is a potential collision risk between two vessels and identify the corresponding risk type. Risk types can include encounter, overtaking, and crossing. Encounter refers to two vessels meeting on opposite or nearly opposite headings, i.e., the angle between their headings is approximately 180 degrees, posing a risk of head-on collision. Overtaking refers to one vessel catching up with and overtaking another vessel from a direction greater than 22.5 degrees aft of its beam; in this case, the overtaking vessel should give way to the overtaken vessel. An intersection refers to a situation where the course of two vessels intersects, and it does not fall under the category of meeting or overtaking. In this case, the responsibility for giving way needs to be determined based on the relative positions and course of the two vessels.

[0068] In this embodiment, the individualized dynamics model combines static attributes such as tonnage, draft, and hull design of a single vessel with real-time speed, load status, and meteorological and hydrological conditions. Through data-optimized physical equations, it accurately captures the unique maneuvering characteristics of the vessel, making trajectory prediction and performance parameter calculations more closely aligned with actual navigation patterns. Furthermore, it constructs a dynamic asymmetric safety boundary, abandoning fixed-form safety domain settings. Based on the vessel's real-time motion state, channel environment, and encounter situation, it adaptively adjusts the range and shape of the safety boundary, making the spatial benchmark for risk assessment more closely match actual collision risk scenarios. This combination not only avoids false alarms caused by insufficient adaptability in traditional models but also reduces missed alarms caused by ignoring individual vessel differences and the influence of the dynamic environment. It is expected to improve warning accuracy by more than 30%. Crew members will no longer need to frequently deal with invalid alarms or question their effectiveness, significantly increasing their trust in the system and making them more willing to take collision avoidance actions based on warning information, thus forming a positive interaction between safety warnings and actual navigation.

[0069] In one embodiment, a specific implementation method for calculating the dynamic safety boundary for each vessel based on the optimal estimated trajectory, theoretical braking distance, and minimum turning radius is also provided, including: Based on dynamic and static data and target detection data, for each radar target, the nearest neighbor AIS target is found in the AIS target list. When a radar target is successfully associated with its nearest neighbor AIS target, the dynamic and static data are bound to the radar target, and fused trajectory data is generated for the bound target. For any two ships, the nearest encounter distance and the time to reach the nearest encounter point are calculated based on the fused trajectory data. The nearest encounter distance is compared with the encounter distance threshold corresponding to each encounter type, and the time to reach the nearest encounter point is compared with the time to reach the nearest encounter point threshold corresponding to each encounter type to determine the collision risk triggering result for the two ships. Based on the optimal estimated trajectory, predicted position, predicted speed, and dynamic safety boundary of the two ships, the relative heading, relative speed, ship type combination, and overlapping area of ​​the safety boundary are determined. The relative heading, relative speed, ship type combination, overlapping area, nearest encounter distance, and time to reach the nearest encounter point are input into the machine learning model to obtain the collision probability value of the two ships. Based on the collision risk triggering result and the collision probability value, the collision risk result between the two ships is determined.

[0070] In this embodiment, the collision risk triggering result indicates whether the two vessels have triggered a collision risk or not. Please refer to [link to relevant documentation]. Figure 7As shown, taking millimeter-wave radar as an example, after acquiring the dynamic and static data (AIS data) of the AIS base station and the target detection data of the radar, in order to avoid the defects of single sensor data, such as the existence of signal blind spots in AIS and random errors in radar data, a fusion algorithm based on nearest neighbor association can be used to fuse data from multiple sources to make up for the defects of low AIS data update rate and missing radar target identity, and generate a fused trajectory with both high accuracy and identity information.

[0071] Specifically, in the process of data fusion processing using a nearest neighbor-based association algorithm, spatial association is first performed. For each radar point cloud target detected by the radar, targets that meet the preset spatial distance threshold are selected from the acquired AIS target list. The nearest neighbor AIS target is identified to ensure that the associated targets are the same vessel; and identity binding is performed when the radar target... With AIS target After a successful match, Static attribute data (identity, ship type, gross tonnage, etc.) gives radar targets This system addresses the limitation of radar in identifying targets, providing a basis for subsequent differentiated risk assessment. It also performs trajectory complementation, generating fused trajectory data for the bound targets. This fused trajectory data includes position, speed, and heading. The position and speed information primarily utilizes frequently updated radar data to ensure real-time trajectory accuracy, while AIS data is used to periodically correct random errors in radar measurements, improving trajectory accuracy. Through data-level fusion, the system not only solves the identification blind spot caused by the lack of AIS for small inland waterway vessels but also optimizes the accuracy of single-sensor data, providing high-quality data support for subsequent risk assessment.

[0072] Furthermore, after acquiring the fused trajectory data, the real-time position, real-time speed, and real-time heading information of the two ships are extracted from the fused trajectory data. Taking one ship as the origin of the reference coordinate system, the position information and velocity vector of the other ship are transformed to obtain its relative position vector and relative velocity vector in the reference coordinate system. Through relative motion analysis, the projection component of the relative velocity vector of the two ships in the direction perpendicular to the relative position vector is calculated. When this projection component is zero, it indicates that the two ships are closest, and this distance is the closest meeting distance (DCPA) between the two ships. At the same time, the projection component of the relative velocity vector in the direction of the relative position vector is calculated, and the absolute value of the projection component is divided by the magnitude of the relative position vector to obtain the time to reach the nearest meeting point (TCPA). The nearest meeting distance is the expected closest distance between the two ships under the current motion state, and the time to reach the nearest meeting point is the time required for the two ships to reach the position corresponding to the nearest meeting distance (DCPA).

[0073] For any two ships and By combining the COLREGs rules with the actual encounter geometry of inland waterways, differentiated encounter distance thresholds can be set for different encounter types (face-to-face, overtaking, and crossing). and arrival time threshold For example, for situations with higher risk, the corresponding encounter distance threshold should be set more strictly; for situations where the situation is relatively controllable, the encounter distance threshold can be appropriately relaxed; and clear risk triggering conditions should be set.

[0074] If and only if the nearest encounter distance between the two ships is... < (Corresponding to the type of encounter) and the time of arrival at the nearest encounter point. < When the encounter type is specified, a rule-based collision risk is determined between the two vessels; otherwise, no collision risk exists. By comparing the nearest encounter distance with the encounter distance threshold corresponding to each encounter type, and comparing the time to reach the nearest encounter point with the time to reach the encounter point threshold corresponding to each encounter type, the collision risk triggering result of the two vessels is determined. This transforms traditionally vague regulatory requirements into quantifiable mathematical conditions, ensuring compliance of the assessment and avoiding misjudgments caused by relying solely on thresholds, thus adapting to the risk characteristics of different encounter scenarios.

[0075] Understandably, inland waterways present numerous complex encounter scenarios, such as multiple vessels crossing each other or meeting in narrow channels. Traditional rule-based judgments struggle to fully cover these ambiguous scenarios, easily leading to missed detections. Therefore, it is necessary to combine machine learning models for collaborative assessment of collision risks to compensate for the limitations of rule-based judgments and improve the system's adaptability to complex scenarios. The aforementioned machine learning models can be random forests, CNN models, or XGBoost binary classifiers.

[0076] After determining the optimal estimated trajectory, predicted position, predicted speed, and dynamic safety boundary for two vessels, the real-time headings of the two vessels are first extracted from the optimal estimated trajectory, and the relative heading is obtained by calculating the difference between the heading angles of the two vessels. The velocity vectors of the two vessels are extracted from the predicted speed, and the relative speed is obtained through vector subtraction. The vessel type information of the two vessels is extracted from the dynamic and static data of the vessels and combined to form a vessel type combination. The dynamic safety boundaries of the two vessels are then projected onto the same geographic coordinate system, and the intersection area of ​​the two elliptical regions is calculated based on the elliptical geometric parameters of each dynamic safety boundary. This intersection area is the overlapping area of ​​the safety boundaries. The elliptical geometric parameters include the center coordinates, major axis length, minor axis length, and heading angle of the ellipse. For example, a vessel type combination could be a 10,000-ton cargo ship and a 1,000-ton barge.

[0077] relative heading Relative velocity Ship type combination Overlapping area Recently, we will encounter distance And the time it will take to reach the nearest point The feature vectors, used as input to the machine learning model, undergo binary classification to obtain the collision probability values ​​of the two ships. The range of values ​​for this collision probability is: In this embodiment, by introducing the overlapping area of ​​relative heading, relative speed, ship type combination, and dynamic safety boundary, the collision probability value can be determined more accurately. Specifically, by introducing the overlapping area... It can directly reflect the degree of conflict in the safety range of the two ships, enabling the model to perceive the safety margin at the physical level, rather than relying solely on kinematic parameters.

[0078] During training, the aforementioned machine learning model can begin by collecting historical navigation data, including numerous samples of safe passage scenarios and collision / high-risk scenarios. Corresponding parametric features are extracted from these samples, including the closest encounter distance (DCPA), time to nearest encounter (TCPA), relative heading, relative speed, vessel type combination, and the overlapping area of ​​dynamic safety boundaries. These parametric feature data undergo preprocessing such as cleaning, normalization, and feature selection to remove noise and redundant information, resulting in a processed dataset. The processed dataset is then proportionally divided into training and testing sets. Algorithms suitable for classification tasks, such as Random Forest or XGBoost, are selected, with collision risk results (risky / no risk) as labels. The initial model is trained using the training set data to obtain the output result, and the initial model parameters are iteratively optimized to minimize the error between the output result and the label result. Finally, the model performance is evaluated using the testing set data, and the model is further optimized by adjusting hyperparameters to ensure good generalization ability on unknown data, thus obtaining a machine learning model capable of effectively predicting collision risks.

[0079] After obtaining the collision probability value and collision risk triggering result, the results of the rule engine and machine learning are integrated to build a hybrid intelligent judgment system with "rules as the main approach and machine learning (ML) as the auxiliary approach". This system not only ensures the compliance and interpretability of the collision risk results, but also takes into account the recognition ability of complex scenarios.

[0080] The aforementioned collision risk results include: risk type; based on the collision risk triggering result and collision probability value, the collision risk result between the two vessels is determined, including: When the collision risk trigger result indicates that two ships have triggered a collision risk, the collision risk result is determined to be that two ships have a collision risk; when the collision risk trigger result indicates that two ships have not triggered a collision risk, and the collision probability value is greater than a preset probability threshold, the collision risk result is determined to be that two ships have a collision risk; in the case of two ships having a collision risk, the risk type of the two ships is identified according to the relative heading and relative speed, according to a preset strategy; the risk types include encounter, overtaking, and crossing.

[0081] Specifically, if the collision outcome indicates that the two vessels are at risk of collision, thus meeting the threshold conditions for DCPA and TCPA, then it can be directly determined that the two vessels are at risk of collision. In this case, the preset rules serve as the core basis to ensure that the decision complies with navigation regulations and avoids missed risk assessments due to biases in the machine learning model. When the collision outcome indicates that the two vessels are not at risk of collision, it is necessary to further judge based on the collision probability value output by the machine learning model, comparing the collision probability value with the preset probability threshold. In comparison, when the collision probability value is higher than the preset probability threshold, it is determined that there is a risk of collision between the two ships. At this time, the blind spots of the rules are supplemented by machine learning results to cover complex scenarios that the rules cannot accurately adapt to, such as cross encounters under the interference of multiple ships.

[0082] In cases where there is a risk of collision between two vessels, the risk type is determined according to a preset strategy based on the relationship between the risk type and the heading angle. Based on the relative heading angles of the two vessels, if their headings are opposite or nearly opposite (i.e., the relative heading angle is around 180 degrees), the risk type is determined to be a head-on collision. If their headings are the same or nearly the same, and one vessel overtakes the other from a direction greater than 22.5 degrees aft of its beam, the risk type is determined to be an overtaking collision. If their headings intersect, but the situation is neither a head-on collision nor an overtaking collision, the risk type is determined to be a crossing collision.

[0083] In this embodiment, the system makes judgments based on COLREGs rules and collaborates with machine learning models to ensure compliance with navigation regulations and accurately address complex encounter scenarios in inland waterways. This provides a precise and reliable basis for generating personalized collision avoidance suggestions, improving robustness, adaptability to complex scenarios, and strong generalization ability. It significantly enhances the robustness and reliability of risk assessment, avoiding the limitations of single assessment methods. Furthermore, by introducing machine learning algorithms and learning from massive amounts of historical navigation data, it can capture and identify potential risk patterns in complex multi-vehicle interaction scenarios that are difficult to describe with fixed rules. For example, in busy waterways where multiple vessels converge, traditional threshold judgment methods are rather general and one-sided when their trajectories intertwine. In contrast, this application uses a machine learning model that, based on its deep understanding of complex patterns, can predict the future intentions of each vessel and make more reasonable and comprehensive decisions. This hybrid strategy makes the system's intelligence level far exceed that of traditional methods that rely solely on thresholds, achieving a leap from compliant operation to intelligent decision-making.

[0084] Step S206: Based on the collision risk results, generate a recommended safe speed and graded early warning information for the target vessel and send it to the corresponding vessel terminal; the target vessel is a vessel with collision risk.

[0085] The aforementioned tiered early warning information includes the warning information content, warning level, risk type, collision avoidance advice, safety confidence level, and a timestamp field. The warning information content can include information corresponding to the three levels: "Alert," "Warning," and "Emergency." The warning level is a classification based on the urgency and severity of the risk. The risk type refers to the specific situation when the two vessels meet, such as head-on collision, overtaking, or crossing. Collision avoidance advice is specific operational guidance provided to the crew by the system based on the current situation, such as "Please immediately reduce speed and turn right to avoid the collision." The safety confidence level indicates the reliability or probability of the system's risk prediction result. The timestamp field records the specific date and time the warning information was generated for traceability and analysis.

[0086] In one embodiment, a specific implementation method for generating a safe recommended speed for a target vessel based on collision risk results is also provided, the method comprising: A feasible speed range is obtained. Using the target vessel's current speed as the initial value, a search is performed within the feasible speed range to determine the optimal candidate speed, which is then used as the recommended safe speed. The optimal candidate speed satisfies the following conditions: the nearest encounter distance to the target vessel is greater than or equal to a safe encounter distance threshold; the time to reach the nearest encounter point is greater than or equal to a safe encounter arrival time threshold; and the difference from the current speed is minimized. Based on the target vessel's current speed, nearest encounter distance, time to reach the nearest encounter point, recommended safe speed, risk type, and collision probability value, a graded alarm message is generated, and the recommended safe speed and graded alarm message are sent to the corresponding vessel terminal of the target vessel.

[0087] It should be noted that the above-mentioned feasible speed range It can be customized according to actual needs, including the lower bound and upper bound of feasible speed. The minimum speed prescribed for the waterway, and the upper limit of the feasible speed. This is the maximum permissible speed calculated based on the ship's dynamics model, which meets the safe braking distance requirements under current conditions. Recommended safe speed. The determination process is an optimization solution process based on a safety model. This process elevates the speed recommendation from a simple query operation to an optimization problem based on constraints, ensuring the scientific nature and optimality of the recommended value.

[0088] Specifically, in generating a safe recommended speed, the process can begin by using the target vessel's current speed as the initial value. Within a preset feasible speed range, a series of candidate speeds are generated at a certain step size, which can be customized as needed, for example, 0.1. Then, for each candidate speed, the trajectory of the target vessel at that speed is simulated, and combined with the predicted trajectory of another vessel, the nearest encounter distance and the time to reach the nearest encounter point are recalculated. Finally, all vessels meeting the "nearest encounter distance" condition are selected. Greater than or equal to the safe encounter distance threshold "And it will take some time to reach the nearest destination." Greater than or equal to the safety threshold will encounter a time threshold. Candidate speeds are selected to form a set of candidate speeds that meet safety conditions. Finally, the candidate speed with the smallest difference from the target vessel's current speed is selected from this set and determined as the optimal candidate speed, which is then used as the final recommended safe speed. Output.

[0089] Specifically, based on the target vessel's current speed, nearest encounter distance, time to nearest encounter point, recommended safe speed, risk type, and collision probability value, a tiered alarm system is generated, and the recommended safe speed and tiered warning information are sent to the target vessel's corresponding terminal, including: Based on the target vessel's current speed, nearest encounter distance, and time to nearest encounter point, a warning message is generated. Specifically, when the target vessel's current speed is above the safe speed limit and there is no collision risk with any other vessel, a warning message of the "notice" level is generated. The safe speed limit is calculated based on dynamic safety boundaries and channel conditions. When the target vessel faces a collision risk, and the nearest encounter distance is below the warning-level encounter distance threshold but not below the emergency-level encounter distance threshold, and the time to nearest encounter point is below the warning-level encounter arrival time threshold but not below the emergency-level encounter arrival time threshold, a warning message of the "warning" level is generated. When the target vessel faces a collision risk, the nearest encounter distance is below the emergency-level encounter distance threshold, and the time to nearest encounter point is below the emergency-level encounter arrival time threshold, or when it is detected that the crew has not taken effective collision avoidance actions within a preset time after the warning-level warning is issued, an emergency-level warning message is generated.

[0090] A comprehensive assessment is conducted on the nearest encounter distance, the time to reach the nearest encounter point, and the collision probability value to generate a safety confidence level. Collision avoidance advice and timestamp fields are obtained. The warning information content, warning level, risk type, collision avoidance advice, safety confidence level, recommended safe speed, and timestamp fields are encapsulated into a ship safety service message and sent to the ship terminal corresponding to the target ship.

[0091] Specifically, in determining the content of the warning information, when the target vessel's current speed... Higher than its dynamic security boundary The upper limit of safe speed calculated based on waterway conditions However, it has not yet posed a collision risk to any vessel, generating a warning message. The corresponding message content is, for example, "Current speed is too high; it is recommended to reduce speed to..." "Should be kept below the section to ensure safety" and this should be displayed via a gentle visual cue on the ship's terminal, such as in blue text.

[0092] When there is a collision risk to the target vessel and the nearest encounter distance is below the warning-level encounter distance threshold. And not lower than the emergency-level encounter distance threshold If the time to reach the nearest warning point is lower than the warning level, a time threshold will be encountered. And it will encounter a time threshold if it is not lower than the emergency level. When this occurs, the generated warning information is at the warning level; the corresponding information content includes at least: "[Warning] [Risk Type] Risk! Target vessel: MMSI, [Recommendation]: Collision avoidance recommendation", and is displayed through the shipboard terminal with prominent visual and clear audio prompts, such as by flashing yellow.

[0093] When a target vessel poses a collision risk, the nearest encounter distance is below the emergency-level encounter distance threshold, or the time to reach the nearest encounter point is below the emergency-level encounter time threshold, or when it is detected that the crew has not taken effective collision avoidance actions within a preset time after the warning-level warning is issued, an emergency-level warning message is generated. The corresponding message content includes at least: "[Emergency] [Risk Type] Collision Approaching! Target Vessel: MMSI, Immediate Execution: Emergency Collision Avoidance Action", and is issued through the shipboard terminal with a strong visual and high-decibel, high-frequency audible and visual alarm, such as a red full-screen flashing display.

[0094] Furthermore, collision avoidance recommendations can be generated by an expert system based on COLREGs rules, ship handling characteristics, and encounter geometry. These recommendations include, but are not limited to: 1) "Turn to Stare". 1) "Decelerate to"; 2) "Decelerate to" Section; 3) "combined action: speed up to section X and turn right to Y degree".

[0095] Recommended speed for safety An additional security confidence level (CCL) is added: This CCL is based on , , and environmental data The comprehensive assessment results are divided into three key indicators: Distance to Nearest Encounter (DCPA), Time to Nearest Encounter (TCPA), and Collision Probability. These indicators are then quantified and scored separately. For example, the closer the DCPA is to the safety threshold or the shorter the TCPA, the lower the score; conversely, the higher the collision probability, the higher the score. Subsequently, based on the importance of each indicator, weight values ​​are determined, for example, through expert experience or historical data. The scores of the three indicators are then integrated into a comprehensive risk score using a weighted summation or other fusion algorithm based on these weight values. Finally, the comprehensive risk score is compared with a preset safety confidence level classification standard (such as high, medium, and low). The corresponding safety confidence level is determined based on the score's range, where a higher comprehensive risk score corresponds to a lower safety confidence level, and vice versa.

[0096] After obtaining collision avoidance recommendations and safety confidence levels, the following information will be used: vessel type MMSI, warning information content, warning level, risk type, collision avoidance recommendations, safety confidence level, and recommended safe speed. The system encapsulates the timestamp field into a Ship Safety Service Message (SSPM) and sends it to the corresponding ship terminal of the target vessel. This SSPM defines a unified data exchange protocol within the system and between the system and the ship terminal, ensuring the accuracy and resolvability of information transmission and forming the basis for system interconnection.

[0097] Furthermore, the shipborne terminal's human-machine interface (HMI) also provides a "cooperative collision avoidance" mode: when two or more vessels receive collision avoidance advice simultaneously, the system can send coordinated collision avoidance commands to them, for example, suggesting one vessel accelerate while the other decelerates, to avoid new risks caused by the two vessels taking opposite actions. The ship terminal also has offline computing capabilities, enabling it to perform basic data processing, thus ensuring basic safety during communication interruptions and achieving synergistic optimization of macro-efficiency and micro-safety.

[0098] In this embodiment, by using the target vessel's current speed as the initial value, a search is performed within the feasible speed range to accurately determine the recommended safe speed. Furthermore, the safety confidence level and graded warning information are determined, enabling crew members to quickly understand the risk level and the system's control. This significantly improves the operability of the guidance information and crew acceptance, ensuring that safety measures are effectively implemented.

[0099] This application provides a ship collision early warning method. Compared with existing technologies, this solution integrates the ship's dynamic and static data with target detection data and uses Kalman filtering to achieve short-term trajectory prediction. This allows for a more comprehensive capture of the ship's real-time and future motion states, providing data guidance for subsequent risk assessment and avoiding misjudgments caused by single sensor data bias. Furthermore, by leveraging a trained ship dynamics model and combining the ship's own dynamic and static data, it outputs personalized theoretical braking distances and minimum turning radii. This overcomes the limitations of traditional technologies that rely solely on fixed thresholds, fully adapting to the differences in braking performance and collision avoidance requirements of ships of different tonnages, such as 10,000-ton cargo ships and 1,000-ton barges, fundamentally improving the adaptability of risk assessment. Additionally, based on the optimal estimated trajectory and theoretical braking distance... By establishing a minimum turning radius and constructing an asymmetric elliptical dynamic safety boundary centered on the ship's hull, the system accurately delineates the safety risk areas that change with the ship's motion state. Compared to traditional fixed geometric boundaries, this approach better reflects the risk distribution patterns in actual navigation. Furthermore, through collaborative analysis using pre-defined rules and machine learning models, the system effectively reduces false alarms and missed alarms, significantly improving the accuracy of risk identification while adapting to complex inland waterway encounter scenarios. Finally, based on the risk results, the system generates recommended safe speeds and tiered warning information, transforming abstract risk judgments into concrete and actionable operational guidelines sent to the ship's terminal. This facilitates rapid and accurate collision avoidance decisions by the crew, significantly improving the timeliness and scientific rigor of collision avoidance operations, enhancing the accuracy of collision warnings, and ensuring the safe and efficient operation of inland waterway transportation.

[0100] Based on the same inventive concept, this application also provides a method for implementing the aforementioned ship collision warning device. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations of one or more ship collision warning device embodiments provided below can be found in the limitations of the ship collision warning method described above, and will not be repeated here.

[0101] In one exemplary embodiment, such as Figure 8 As shown, a ship collision warning device is provided, the device comprising: The acquisition module 510 is used to acquire dynamic and static data and target detection data of each ship in the waterway; The processing module 520 is used to perform spatiotemporal registration and fusion of dynamic and static data and target detection data, generate the optimal estimated trajectory for each ship, use a Kalman filter to perform short-term trajectory prediction based on the optimal estimated trajectory, and output the predicted position and predicted speed at multiple future times. The first determining module 530 is used to input dynamic and static data into the trained ship dynamics model to obtain the theoretical braking distance and minimum turning radius of the ship under the current load and waterway conditions. The second determining module 540 is used to calculate the dynamic safety boundary for each ship based on the optimal estimated trajectory, theoretical braking distance and minimum turning radius; the dynamic safety boundary is an asymmetric elliptical region centered on the ship's hull that changes over time. The risk identification module 550 is used to predict collision risks between ships by using the optimal estimated trajectory, predicted position, predicted speed and dynamic safety boundary through preset rules and machine learning models. The collision warning module 560 is used to generate a safe recommended speed and graded warning information for the target vessel based on the collision risk results and send it to the corresponding vessel terminal; the target vessel is a vessel with a collision risk.

[0102] As an optional implementation, the first determining module 530 is specifically used for: Obtain meteorological and hydrological data on the current environment of the ship; Meteorological and hydrological data, ship tonnage, real-time draft, real-time speed, and channel conditions are used as feature vectors and input into the braking distance model to output the theoretical braking distance. Meteorological and hydrological data, real-time speed, and structural attribute parameters are input into the minimum turning radius model to obtain the minimum turning radius of the ship under the current load and channel conditions. The structural attribute parameters include: ship length and ship width.

[0103] As an optional implementation, the ship dynamics model is constructed through the following steps: The system acquires multi-source operational data of the vessel and preprocesses the data to obtain processed data. The preprocessing includes data cleaning and data labeling. The processed data includes: vessel static parameters, meteorological and hydrological data samples, actual braking distance and actual minimum turning radius. The vessel static parameters include: vessel tonnage samples, draft samples, initial speed samples, channel condition samples, and structural attribute samples. Meteorological and hydrological data samples, ship tonnage samples, draft samples, initial speed samples, and channel condition samples are input into the initial braking distance model to obtain the predicted braking distance. Meteorological and hydrological data samples, initial speed samples, and structural attribute samples are input into the initial minimum turning radius model to obtain the predicted minimum turning radius. A loss function is constructed based on the predicted braking distance and the actual braking distance, as well as based on the predicted minimum turning radius and the actual minimum turning radius; By minimizing the loss function, machine learning algorithms are used to train the model parameters of the initial braking distance model and the initial minimum turning radius model to obtain the ship dynamics model.

[0104] As an optional implementation, the second determining module 540 is specifically used for: The real-time turning rate of the ship is determined based on the optimal estimated trajectory; Calculate the basic length of the major axis of the elliptical region based on real-time speed and theoretical braking distance; Calculate the basic length of the minor axis of the elliptical region based on the minimum turning radius and real-time turning rate; Based on meteorological and hydrological data, the basic lengths of the major axis and minor axis are corrected to obtain the corrected lengths of the major axis and minor axis. Based on the positive or negative magnitude of the real-time steering rate and the minor axis correction length, calculate the left and right minor axis lengths of the elliptical region; The ship's center of mass is used as the coordinates of the center point of the elliptical region, the ship's heading angle is used as the major axis direction angle of the elliptical region, and the major axis correction length, the left minor axis length, and the right minor axis length are used as the major and minor axes of the ellipse to determine the dynamic safety boundary.

[0105] As an optional implementation, the risk identification module 550 is specifically used for: Based on dynamic and static data and target detection data, for each radar target, the nearest neighbor AIS target is found in the AIS target list; When a radar target is successfully associated with the nearest neighboring AIS target, dynamic and static data are bound to the radar target, and fused trajectory data is generated for the bound target. For any two ships, calculate the nearest encounter distance and the time to reach the nearest encounter point based on the fused trajectory data; The nearest encounter distance is compared with the encounter distance threshold corresponding to each encounter type, and the time to reach the nearest encounter point is compared with the time to reach the encounter point corresponding to each encounter type to determine the collision risk triggering result of the two ships. Based on the optimal estimated trajectories, predicted positions, predicted speeds, and dynamic safety boundaries of the two ships, the relative headings, relative speeds, ship type combinations, and overlapping areas of the safety boundaries of the two ships are determined. The relative heading, relative speed, combination of ship types, overlap area, nearest encounter distance, and time to reach the nearest encounter point are input into the machine learning model to obtain the collision probability value of the two ships. Based on the collision risk triggering result and the collision probability value, the collision risk result between the two ships is determined.

[0106] As an optional implementation, the risk identification module 550 is also used for: When the collision risk trigger result indicates that two ships have triggered a collision risk, the collision risk result is determined to mean that there is a collision risk between the two ships. When the collision risk trigger result indicates that the two ships have not triggered a collision risk, if the collision probability value is greater than the preset probability threshold, the collision risk result is determined to be that the two ships have a collision risk. In the event of a collision risk between two vessels, the risk type of the two vessels is identified according to their relative headings and relative speeds, following a pre-defined strategy. The risk types include head-on collision, overtaking, and crossing.

[0107] As an optional implementation, the collision warning module 560 is specifically used for: Obtain the feasible speed range; the feasible speed range includes the lower bound of the feasible speed and the upper bound of the feasible speed. The lower bound of the feasible speed is the minimum speed specified by the waterway, and the upper bound of the feasible speed is the maximum allowable speed calculated based on the ship dynamics model and meeting the safe braking distance under the current conditions. Using the target vessel's current speed as the initial value, a search is conducted within the feasible speed range to determine the optimal candidate speed, which is then used as the recommended safe speed. The optimal candidate speed satisfies the following conditions: the nearest encounter distance to the target vessel is greater than or equal to the safe encounter distance threshold, the time to reach the nearest encounter point is greater than or equal to the safe encounter arrival time threshold, and the difference from the current speed is minimized. Based on the target vessel's current speed, nearest encounter distance, time to nearest encounter point, recommended safe speed, risk type, and collision probability value, a graded alarm message is generated, and the recommended safe speed and graded warning message are sent to the corresponding vessel terminal.

[0108] As an optional implementation, the collision warning module 560 is also used for: Based on the target vessel's current speed, nearest encounter distance, and time to nearest encounter point, a warning message is generated. Specifically, when the target vessel's current speed is above the safe speed limit and there is no collision risk with any other vessel, a warning message of the "notice" level is generated. The safe speed limit is calculated based on dynamic safety boundaries and channel conditions. When the target vessel faces a collision risk, and the nearest encounter distance is below the warning-level encounter distance threshold but not below the emergency-level encounter distance threshold, and the time to nearest encounter point is below the warning-level encounter arrival time threshold but not below the emergency-level encounter arrival time threshold, a warning message of the "warning" level is generated. When the target vessel faces a collision risk, the nearest encounter distance is below the emergency-level encounter distance threshold, and the time to nearest encounter point is below the emergency-level encounter arrival time threshold, or when it is detected that the crew has not taken effective collision avoidance actions within a preset time after the warning-level warning is issued, an emergency-level warning message is generated. A comprehensive assessment is conducted on the nearest encounter distance, the time to reach the nearest encounter point, and the collision probability value to generate a safety confidence level. Obtain the collision avoidance advice and timestamp fields, encapsulate the warning information content, warning level, risk type, collision avoidance advice, safety confidence level, recommended safe speed, and timestamp fields into a ship safety service message, and send it to the ship terminal corresponding to the target ship.

[0109] The ship collision warning device provided in this application embodiment, by fusing the ship's dynamic and static data with target detection data and performing Kalman filtering to achieve short-term trajectory prediction, can more comprehensively capture the ship's real-time state and future motion state, providing data guidance for subsequent risk assessment and avoiding misjudgments caused by single sensor data bias; and by using a trained ship dynamics model, combined with the ship's own dynamic and static data, it outputs personalized theoretical braking distance and minimum turning radius, which can overcome the limitations of traditional technologies that only use fixed thresholds for judgment, and fully adapt to the differences in braking performance and collision avoidance operation requirements of ships of different tonnages, such as 10,000-ton cargo ships and 1,000-ton barges, fundamentally improving the adaptability of risk assessment; and based on the optimal estimated trajectory, theoretical braking distance and By constructing an asymmetric elliptical dynamic safety boundary centered on the hull using the minimum turning radius, the system accurately delineates the safety risk areas that change with the ship's motion state. Compared to traditional fixed geometric boundaries, this approach better reflects the risk distribution patterns in actual navigation. Furthermore, through the collaborative analysis of preset rules and machine learning models, the system effectively reduces false alarms and missed alarms, significantly improving the accuracy of risk identification while adapting to complex inland waterway encounter scenarios. Finally, based on the risk results, the system generates recommended safe speeds and tiered warning information, transforming abstract risk judgments into concrete and actionable operational guidelines sent to the ship's terminal. This facilitates rapid and accurate collision avoidance decisions by the crew, significantly improving the timeliness and scientific rigor of collision avoidance operations, enhancing the accuracy of collision warnings, and ensuring the safe and efficient operation of inland waterway transportation.

[0110] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 9 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores video tag processing data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by the processor, the computer program implements a ship collision warning method.

[0111] Those skilled in the art will understand that Figure 9The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0112] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0113] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0114] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0115] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0116] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0117] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0118] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0119] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A ship collision early warning method, characterized in that, The ship collision early warning method includes: Acquire dynamic and static data and target detection data of each vessel in the waterway; The dynamic and static data and the target detection data are spatiotemporally registered and fused to generate an optimal estimated trajectory for each ship. Based on the optimal estimated trajectory, a Kalman filter is used to predict the short-term trajectory and output the predicted position and predicted speed at multiple future times. The dynamic and static data are input into the trained ship dynamics model to obtain the theoretical braking distance and minimum turning radius of the ship under the current load and waterway conditions. Based on the optimal estimated trajectory, the theoretical braking distance, and the minimum turning radius, a dynamic safety boundary is calculated for each vessel; the dynamic safety boundary is an asymmetric elliptical region centered on the vessel's hull that changes over time. The optimal estimated trajectory, predicted position, predicted speed, and dynamic safety boundary are used to predict collision risk and identify collision risk results between ships through preset rules and machine learning models. Based on the collision risk results, a safe recommended speed and graded early warning information are generated for the target vessel and sent to the corresponding vessel terminal; the target vessel is a vessel with a collision risk.

2. The ship collision early warning method according to claim 1, characterized in that, The dynamic and static data include: ship tonnage, structural attribute parameters, real-time speed, channel conditions, and real-time draft; the ship dynamics model includes a braking distance model and a minimum turning radius model. The dynamic and static data are input into the trained ship dynamics model to obtain the theoretical braking distance and minimum turning radius of the ship under the current load and channel conditions, including: Obtain meteorological and hydrological data of the current environment of the vessel; The meteorological and hydrological data, ship tonnage, real-time draft, real-time speed, and channel conditions are used as feature vectors and input into the braking distance model to output the theoretical braking distance. The meteorological and hydrological data, real-time speed, and structural attribute parameters are input into the minimum turning radius model to obtain the minimum turning radius of the ship under the current load and channel conditions. The structural attribute parameters include: ship length and ship width.

3. The ship collision early warning method according to claim 2, characterized in that, The ship dynamics model is constructed through the following steps: Acquire multi-source operational data of the vessel and preprocess the multi-source operational data to obtain processed data; The preprocessing includes: data cleaning and data labeling; the processed data includes: ship static parameters, meteorological and hydrological data samples, actual braking distance and actual minimum turning radius; the ship static parameters include: ship tonnage samples, draft samples, initial speed samples, channel condition samples, and structural attribute samples; The meteorological and hydrological data samples, ship tonnage samples, draft samples, initial speed samples, and channel condition samples are input into the initial braking distance model to obtain the predicted braking distance. The meteorological and hydrological data samples, initial speed samples, and structural attribute samples are input into the initial minimum turning radius model to obtain the predicted minimum turning radius. A loss function is constructed based on the predicted braking distance and the actual braking distance, as well as based on the predicted minimum turning radius and the actual minimum turning radius; By minimizing the loss function, a machine learning algorithm is used to train the model parameters of the initial braking distance model and the initial minimum turning radius model to obtain the ship dynamics model.

4. The ship collision early warning method according to claim 2, characterized in that, Based on the optimal estimated trajectory, the theoretical braking distance, and the minimum turning radius, a dynamic safety boundary is calculated for each vessel, including: Based on the optimal estimated trajectory, the real-time turning rate of the vessel is determined; Based on the real-time speed and the theoretical braking distance, calculate the basic length of the major axis of the elliptical region; Based on the minimum turning radius and the real-time steering rate, calculate the basic length of the minor axis of the elliptical region; Based on the meteorological and hydrological data, the basic length of the major axis and the basic length of the minor axis are corrected to obtain the corrected length of the major axis and the corrected length of the minor axis. Based on the positive or negative magnitude of the real-time steering rate and the minor axis correction length, calculate the left minor axis length and the right minor axis length of the elliptical region; The dynamic safety boundary is determined by taking the ship's center of mass as the coordinates of the center point of the elliptical region, taking the ship's bow angle as the major axis direction angle of the elliptical region, and taking the major axis correction length, the left minor axis length, and the right minor axis length as the major and minor axes of the ellipse.

5. The ship collision early warning method according to claim 1, characterized in that, The optimal estimated trajectory, predicted position, predicted speed, and dynamic safety boundary are used to predict collision risk through preset rules and machine learning models, identifying collision risk results between ships, including: Based on the dynamic and static data and the target detection data, for each radar target, the nearest neighbor AIS target is found in the AIS target list; When the radar target is successfully associated with the nearest neighboring AIS target, the dynamic and static data are bound to the radar target, and fused trajectory data is generated for the bound target; For any two ships, calculate the nearest encounter distance and the time to reach the nearest encounter point based on the fused trajectory data; The nearest encounter distance is compared with the encounter distance threshold corresponding to each encounter type, and the time to reach the nearest encounter point is compared with the time to reach the encounter point threshold corresponding to each encounter type to determine the collision risk triggering result of the two ships. Based on the optimal estimated trajectory, predicted position, predicted speed and dynamic safety boundary of the two ships, the relative heading, relative speed, ship type combination and overlapping area of ​​the safety boundary of the two ships are determined. The relative heading, relative speed, ship type combination, overlap area, nearest encounter distance, and time to reach the nearest encounter point are input into the machine learning model to obtain the collision probability value of the two ships. Based on the collision risk triggering result and the collision probability value, the collision risk result between the two ships is determined.

6. The ship collision early warning method according to claim 5, characterized in that, The collision risk outcome includes: risk type; based on the collision risk triggering outcome and the collision probability value, the collision risk outcome between the two vessels is determined, including: When the collision risk trigger result indicates that two ships have triggered a collision risk, the collision risk result is determined to mean that there is a collision risk between the two ships. When the collision risk trigger result indicates that the two ships have not triggered a collision risk, if the collision probability value is greater than a preset probability threshold, the collision risk result is determined to mean that the two ships have a collision risk. In the event of a collision risk between the two vessels, the risk type of the two vessels is identified according to the relative heading and the relative speed, following a preset strategy; the risk type includes encounter, overtaking, and crossing.

7. The ship collision early warning method according to claim 6, characterized in that, Based on the collision risk results, a safe recommended speed and graded early warning information are generated for the target vessel and sent to the corresponding vessel terminal, including: Obtain a feasible speed range; the feasible speed range includes a lower bound and an upper bound, the lower bound is the minimum speed specified by the waterway, and the upper bound is the maximum permissible speed calculated based on the ship dynamics model and meeting the safe braking distance under the current conditions; Using the current speed of the target vessel as the initial value, a search is performed within the feasible speed range to determine the optimal candidate speed, and the optimal candidate speed is used as the safe recommended speed. The optimal candidate speed satisfies the following conditions: the nearest encounter distance to the target vessel is greater than or equal to the safe encounter distance threshold, the time to reach the nearest encounter point is greater than or equal to the safe encounter arrival time threshold, and the difference from the current speed is minimized. Based on the target vessel's current speed, nearest encounter distance, time to nearest encounter point, recommended safe speed, risk type, and collision probability value, a graded alarm message is generated, and the recommended safe speed and graded alarm message are sent to the vessel terminal corresponding to the target vessel.

8. The ship collision early warning method according to claim 7, characterized in that, The tiered early warning information includes the warning information content, warning level, risk type, collision avoidance suggestions, safety confidence level, and timestamp field; Based on the target vessel's current speed, nearest encounter distance, time to nearest encounter point, recommended safe speed, risk type, and collision probability value, a tiered alarm message is generated, and the recommended safe speed and tiered alarm message are sent to the vessel terminal corresponding to the target vessel, including: Based on the target vessel's current speed, nearest encounter distance, and time to nearest encounter point, a warning message is generated. Specifically, when the target vessel's current speed is higher than the safe speed limit and it does not pose a collision risk with any other vessel, a warning message of the advisory level is generated. The safe speed limit is calculated based on the dynamic safety boundary and channel conditions. When the target vessel faces a collision risk, and the nearest encounter distance is lower than the warning-level encounter distance threshold but not lower than the emergency-level encounter distance threshold, and the time to nearest encounter point is lower than the warning-level encounter arrival time threshold but not lower than the emergency-level encounter arrival time threshold, a warning message of the warning level is generated. When the target vessel faces a collision risk, the nearest encounter distance is lower than the emergency-level encounter distance threshold, and the time to nearest encounter point is lower than the emergency-level encounter arrival time threshold, or when it is detected that the crew has not taken effective collision avoidance actions within a preset time after the warning-level warning is issued, an emergency-level warning message is generated. A comprehensive evaluation is performed on the nearest encounter distance, the time to reach the nearest encounter point, and the collision probability value to generate a safety confidence level. Obtain the collision avoidance advice and timestamp field, encapsulate the warning information content, warning level, risk type, collision avoidance advice, safety confidence level, recommended safe speed and timestamp field into a ship safety service message and send it to the ship terminal corresponding to the target ship.

9. A ship collision warning device, characterized in that, The ship collision warning device includes: The acquisition module is used to acquire dynamic and static data and target detection data of each ship in the waterway; The processing module is used to perform spatiotemporal registration and fusion of the dynamic and static data and the target detection data, generate the optimal estimated trajectory for each ship, and use a Kalman filter to perform short-term trajectory prediction based on the optimal estimated trajectory, and output the predicted position and predicted speed at multiple future times. The first determining module is used to input the dynamic and static data into the trained ship dynamics model to obtain the theoretical braking distance and minimum turning radius of the ship under the current load and waterway conditions. The second determining module is used to calculate the dynamic safety boundary for each ship based on the optimal estimated trajectory, the theoretical braking distance, and the minimum turning radius; the dynamic safety boundary is an asymmetric elliptical region centered on the ship's hull that changes over time. The risk identification module is used to predict collision risks between ships by using the optimal estimated trajectory, predicted position, predicted speed and dynamic safety boundary through preset rules and machine learning models; The collision warning module is used to generate a safe recommended speed and graded warning information for the target vessel based on the collision risk results and send it to the vessel terminal corresponding to the target vessel; the target vessel is a vessel with a collision risk.

10. A computer device, comprising: The memory and processor contain a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the ship collision warning method according to any one of claims 1-8.

Images