Water area path planning method and device and storage medium

By acquiring and analyzing historical data of ship collision damage, deep learning models are constructed to predict the type and degree of ship collision damage, which solves the accuracy of ship collision risk assessment in marine shipping and improves the accuracy of navigation safety and risk assessment.

CN118913278APending Publication Date: 2024-11-08GUANGZHOU UNIVERSITY
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410963602.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the marine shipping industry, how to quickly and accurately predict the types and levels of ship collision damage in order to adopt effective path planning remains a challenge, especially in case of ship collision accidents.

Method used

By obtaining historical data of ship collision damage, determining the characteristic parameters of ship collision damage, and building a deep learning model for training, combining real-time data to calculate the risk results of ship collision damage in order to carry out water path planning.

Benefits of technology

It significantly improves the accuracy of risk assessment and the safety of ship navigation, ensuring that effective emergency measures can be taken in a timely manner when a collision risk occurs, reducing economic losses and improving navigation safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118913278A_ABST
    Figure CN118913278A_ABST
Patent Text Reader

Abstract

The invention relates to a water area path planning method and device and a storage medium, and the method comprises the following steps: obtaining the historical data of ship collision damage when there is a collision risk; determining characteristic parameters of ship collision damage according to the historical data; constructing a deep learning model through the feature parameters and carrying out model training; inputting real-time data of ship navigation into the deep learning model, and calculating to obtain a risk result of ship collision damage; and planning a water area path of ship driving based on the risk result. According to the technical scheme, when the collision risk exists, the damage type and the damage degree of the ship are rapidly and accurately predicted so that effective emergency measures can be taken for water area path planning.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to a water area path planning method, device and storage medium. Background Art

[0002] In today's marine shipping industry, the prediction of ship collision damage risk outcomes is a crucial research area, which is directly related to ship safety, environmental protection and economic loss control. With the growth of global trade and the increase in maritime traffic density, the risk of ship collisions and collisions between ships and other obstacles has also increased. Although existing navigation technology and ship management systems have helped to reduce the probability of collision accidents to a great extent, once there is a risk of a collision, how to quickly and accurately predict the type and extent of damage to the ship in order to take effective path planning is still a challenge. Summary of the invention

[0003] The main purpose of the present invention is to provide a water path planning method, which aims to quickly and accurately predict the damage type and damage degree of a ship so as to take effective emergency measures for path planning.

[0004] A first aspect of the present invention provides a water path planning method, the method comprising the following steps:

[0005] When there is a risk of collision, obtain historical data on ship collision damage;

[0006] Determining characteristic parameters of ship collision damage based on the historical data;

[0007] Constructing a deep learning model through the characteristic parameters and performing model training;

[0008] Inputting the real-time data of ship navigation into the deep learning model to calculate the risk results of ship collision damage;

[0009] The water route planning for the ship is carried out based on the risk results.

[0010] Optionally, the step of "obtaining historical data of ship collision damage" includes:

[0011] The historical data of the ship's operation during the collision damage and the historical data of the water environment in which the ship is located are obtained, wherein the historical data of the ship's operation include the cost data of the degree of damage to the ship, the cost data of obstructions to navigation, and the cost data of the length of the path along which the damaged ship travels.

[0012] Optionally, the step of “determining characteristic parameters of ship collision damage according to the historical data” includes:

[0013] The cost value is calculated by a characteristic cost extraction function, and the characteristic parameters of the ship collision damage are determined according to the cost value; the characteristic cost extraction function is: F = α·L+(1-α)·C+T, α is the weight of the path length cost data, (1-α) is the weight of the obstruction cost data, T is the cost data of the degree of damage to the hull, C is the obstruction cost data, and L is the path length cost data; F represents the cost value.

[0014] Optionally, the step of “building a deep learning model through the feature parameters and performing model training” includes:

[0015] Input the characteristic parameters into the deep learning model, calculate the weight values ​​of each layer of the neural network, and output the predicted value of the ship collision damage;

[0016] Calculating an error value between the predicted value and the historical data, and when the error value is greater than a preset error value, updating the weight values ​​of each layer of the neural network;

[0017] The updated weight value is input into the deep learning model for iterative training until a predetermined number of iterations is reached or the calculated error value is no greater than the preset error value.

[0018] Optionally, in the step of "updating the weight values ​​of each layer of the neural network", the weights are updated using a gradient descent method.

[0019] Optionally, before the step of "obtaining historical data of ship collision damage", the step also includes: obtaining real-time data of ship navigation, and determining whether there is a collision risk between the ship and the obstacle based on the real-time data.

[0020] Optionally, the step of “determining whether there is a collision risk between the ship and the obstacle according to the real-time data” includes:

[0021] Utilizing the real-time data to calculate the closest encounter distance and closest encounter time between the ship and each obstacle in the water environment;

[0022] Determine the obstacle with the greatest collision risk according to the closest approach distance and the closest approach time;

[0023] Using the GJK algorithm to calculate the shortest distance between the ship and the obstacle;

[0024] It is determined whether the shortest distance is less than a preset safety threshold. If so, it is determined that there is a collision risk; otherwise, there is no collision risk.

[0025] Optionally, the step of "calculating the risk result of ship collision damage" includes:

[0026] Passing and calculating the real-time data in each neural network layer of the deep learning model;

[0027] The risk results of the collision damage type and the collision damage degree are outputted based on the calculation of the real-time data in each layer of the neural network.

[0028] The present invention also provides a water area path planning device, the device comprising:

[0029] A data acquisition unit, used to acquire historical data of ship collision damage and real-time data of ship navigation;

[0030] A construction unit, used to determine characteristic parameters of ship collision damage according to the historical data, and to construct a deep learning model through the characteristic parameters and perform model training;

[0031] A calculation unit, used to input the real-time data of ship navigation into the deep learning model to calculate the risk result of ship collision damage;

[0032] An execution unit is used to plan a water path for a ship based on the risk result.

[0033] The present invention also provides a computer storage medium, on which a computer program is stored, and the program is used to implement the steps of the above-mentioned water area path planning method.

[0034] In the water path planning method of the present invention, when there is a collision risk, historical data on ship collision damage and real-time data on ship navigation are obtained, and the ship collision damage feature parameters that are most helpful to the deep learning model in the historical data are extracted through a feature cost extraction function. The ship collision damage feature parameters are used to build and train a deep learning model to obtain a deep learning model for predicting the risk results of the ship collision damage type and the degree of collision damage. The processed real-time data of ship navigation is input into the deep learning model for calculation, thereby outputting the risk results of the ship's damage type and degree of damage, so that effective emergency measures can be taken according to the risk results, and water path planning is carried out during the ship's navigation process, which significantly improves the accuracy of risk assessment and the safety of ship navigation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the structures shown in these drawings without paying creative work.

[0036] Figure 1is a flow chart of an embodiment of a water area path planning method of the present invention;

[0037] Figure 2 It is a flow chart of an embodiment of the step of determining whether there is a collision risk between a ship and an obstacle according to real-time data in the water path planning method of the present invention;

[0038] Figure 3 yes Figure 2 The operation flow chart of the GJK algorithm in step;

[0039] Figure 4 It is a flowchart of a specific implementation of step S130 in the water path planning method of the present invention;

[0040] Figure 5 It is a flowchart of a specific implementation of the step of calculating the risk result of ship collision damage in the water path planning method of the present invention;

[0041] Figure 6 It is a structural schematic diagram of an embodiment of the water area path planning device of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only embodiments of a part of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present application.

[0043] Embodiment 1

[0044] See also Figure 1 The present invention provides a water path planning method, which comprises the following steps:

[0045] S110: when there is a collision risk, obtaining historical data of ship collision damage;

[0046] S120: Determine characteristic parameters of ship collision damage according to historical data of ship collision damage;

[0047] S130: constructing a deep learning model based on the characteristic parameters of ship collision damage and performing model training;

[0048] S140: inputting the real-time data of ship navigation into the deep learning model to calculate the risk result of ship collision damage;

[0049] S150: Plan the vessel's route in the waters based on the risk results.

[0050] In the water path planning method of the present invention, when there is a collision risk, historical data on ship collision damage and real-time data on ship navigation are obtained, and the ship collision damage feature parameters that are most helpful to the deep learning model in the historical data are extracted through a feature cost extraction function. The ship collision damage feature parameters are used to build and train a deep learning model to obtain a deep learning model for predicting the risk results of the ship collision damage type and the degree of collision damage. The processed real-time data of ship navigation is input into the deep learning model for calculation, thereby outputting the risk results of the ship's damage type and degree of damage, so that effective emergency measures can be taken according to the risk results, and water path planning is carried out during the ship's navigation process, which significantly improves the accuracy of risk assessment and the safety of ship navigation.

[0051] It should be noted that, in step S110, this embodiment directly obtains historical data of ship collision damage from the ship's sensors. The sensors may be radar, automatic identification system (Automatic Identification System, AIS), shipborne voyage data recorder (VDR), etc. The historical data of ship collision damage refers to the historical data of the hull operation and the historical data of the water environment during the collision damage process. In addition to the cost data of the degree of hull damage, the cost data of obstruction to navigation, and the cost data of the length of the path of the damaged hull, it also includes image data of the damaged area of ​​the ship. These data are usually taken by the on-site investigation team after the accident or collected using professional equipment such as drones. The image data can intuitively show the degree and location of the damage, which is crucial for subsequent maintenance and accident cause analysis, and can be used for damage assessment after the accident; damage location and degree data, which record in detail the damage to various parts of the ship in the collision accident, help estimate the impact of the accident on the survivability of the ship, and can be used to formulate a more effective emergency response plan; the historical track of the ship, which usually records the real-time position and speed information of the ship through the automatic identification system, and obtains the historical track in combination with GPS data. The historical track of the ship records the navigation information of the ship. Information such as speed, direction and stop points during navigation is very important for analyzing the cause of the accident and the navigation habits of the ship. By analyzing the historical track, high-risk areas and time periods can be identified, so that preventive measures can be taken, which can be used to evaluate the safety of the route and optimize the route design; ship size and performance parameters, including the length, width, draft, deadweight tonnage, main engine power and speed of the ship, these parameters directly affect the maneuverability and seaworthiness of the ship. When an accident occurs, these parameters can help evaluate the performance of the ship in different situations and the extent of possible damage. For example, larger ships may react more slowly in collision avoidance operations, while fast-speed ships may require longer braking distances; chart information, the chart provides detailed information on the depth of the water, channel distribution, obstacles and navigation signs. After the accident, the chart information helps to determine whether the accident was caused by an error in the chart or failure to navigate according to the chart; meteorological conditions, including wind speed, wind direction, visibility, wave height and flow direction. Meteorological conditions have an important impact on navigation safety, and severe weather is often one of the direct causes of accidents. Other factors that may affect navigation include but are not limited to maritime traffic density, ship flow, tides and currents, and the mechanical state of the ship. The data generated when the GJK algorithm is applied is conducive to combining the deep learning model with the GJK algorithm, so that the direction of the output results is consistent, thereby improving the reliability of the output results.

[0052] In step S130, a deep learning model is constructed and trained through the characteristic parameters of ship collision damage; during the training process, the deep learning model continuously learns and iteratively trains the input characteristic parameters of ship collision damage, and learns the meaning of the input historical data for later use in realizing the prediction function, for example, learning the image data of the damaged area of ​​the ship, and learning that when a batch of image data is input, the deep learning model can identify which images and which positions have damage, what the type of damage is, and how much the damage is. Specifically, the process of using the deep learning model to learn the image data of the damaged area of ​​the ship is as follows:

[0053] Collect a large amount of image data of damaged areas of the ship, including different types and degrees of damage;

[0054] Accurately label these data to indicate the specific location of the damage, the type of damage (such as scratches, dents, corrosion, etc.), and the extent of the damage;

[0055] Enhance image data by rotating, scaling, cropping, etc. to improve the generalization ability of the model;

[0056] Divide the collected and processed data set into training set, validation set and test set;

[0057] Choose a suitable deep learning model as a basis, such as a convolutional neural network, because convolutional neural networks are very suitable for processing image data. You can also use a pre-trained convolutional neural network model and use transfer learning techniques to adjust the model to suit the current task.

[0058] Depending on specific needs, network layers can be added or modified, such as adding more convolutional layers or fully connected layers to better capture damage characteristics.

[0059] Design feature cost extraction function: F = α·L+(1-α)·C+T(1);

[0060] Among them, α is the weight of the path length cost, (1-α) is the weight of the obstacle cost, T is the cost of the damage degree of the aircraft, C is the obstacle cost, L is the path length cost, and F is the cost value.

[0061] Among them, the cost function of the body damage degree is:

[0062]

[0063] The damage cost function is used to evaluate the type, degree and probability of damage that a ship may suffer during its mission. N is the number of statistical types of different damage types, a i is the influence coefficient of the i-th damage type on continued navigation, b iis the severity of damage caused by the i-th damage type, δ is the corresponding weight, is the probability of damage occurring.

[0064] The obstruction cost function is:

[0065]

[0066] Among them, d ij is the distance between point i on the ship's path and obstacle j, and N is the total number of obstacles.

[0067] The path length cost function is:

[0068]

[0069] Among them, (x i ,y i ) and (x i+1 ,y i+1 ) are the coordinates of adjacent path points on the path.

[0070] Select the Adam optimizer or the RMSprop optimizer, which usually provide better convergence speed and stability.

[0071] Based on the evaluation results, the model is adjusted and optimized to improve the accuracy and reliability of damage assessment and potential impact discovery. For example, regularization techniques such as L1, L2 regularization or Dropout are applied to reduce the risk of overfitting of the model and enhance its generalization ability.

[0072] Experimentally adjust hyperparameters such as learning rate, batch size, and number of training epochs to optimize model performance.

[0073] The model performance was evaluated on the validation set and the test set using metrics such as accuracy, recall, and F1 score.

[0074] Analyze the model's performance on specific types of damage identification to determine if further optimization is needed.

[0075] Using the model in real-world applications, incoming ship images are analyzed in real time to identify and assess damage.

[0076] Monitor the performance of the model in real applications and collect samples that are misclassified or have poor performance.

[0077] Regularly retrain or fine-tune the model with newly collected data to ensure the accuracy and effectiveness of the model.

[0078] Focus on the latest deep learning technologies and research results, and continuously improve model structures and training methods.

[0079] Through the above steps, a deep learning model can be implemented to effectively identify the location, type and extent of damage in ship images. This not only improves the efficiency of damage prediction, but also greatly enhances the ability of preventive maintenance and accident investigation.

[0080] In S140, the real-time data of the ship's navigation is input into the deep learning model to calculate the risk result of the ship's collision damage;

[0081] Collect real-time data of ship navigation, including real-time data of the water environment: hydrological and meteorological information, ship dynamics, static obstacles, etc.; real-time data of ship operation: ship position, speed, heading, etc. Preprocess these data, such as cleaning and normalization, so that they can be used to input deep learning models, calculate the risk results of ship collision damage, and realize the prediction function and the implementation of the GJK algorithm.

[0082] Different collision scenarios are simulated in the model through the input real-time data, and the impact of different factors (such as speed, angle, hull structure) on the degree of damage is evaluated. The output of the deep learning model is combined to predict the possible type and degree of damage. The multi-label classification method is used to allow the model to output the prediction of damage type and severity at the same time. Through the input real-time data, the status information of the ship and dynamic obstacles can be updated in real time to adapt to the changing environment.

[0083] Based on the prediction of the type and severity of ship damage output by the model, appropriate emergency measures can be taken in a timely manner to ensure maximum safety of the ship's navigation and maximum mission completion.

[0084] In S150, a water route planning for the ship is performed based on the risk result.

[0085] Specifically, according to the risk results, understand the type of damage such as structural damage, mechanical damage, etc. and the degree such as minor, moderate, and severe. Then automatically start the emergency response procedure, such as adjusting the route, reducing the speed, closing the damaged area, etc. Using advanced algorithms and chart data, intelligently plan the safest and most economical route, taking into account the damage, weather, sea conditions, and traffic conditions. Continuously monitor the status of the ship and the external environment, dynamically assess risks, and adjust the route as needed. And communicate with the remote control center in real time to transmit data and receive instructions. Comply with maritime regulations to ensure that path planning complies with the requirements of the International Maritime Organization and other relevant maritime regulations. Consider marine environmental factors on the route, such as tides, currents, water depth, etc., to avoid further damage. According to the damage, select the nearest safe port or maintenance facility as an emergency stop. Consider possible energy loss or reduced efficiency, optimize energy use, and ensure that the ship can reach its destination safely. Use simulation tools to predict possible route conditions, including encounters with other ships, collision avoidance, etc. Design redundant control systems and navigation systems to ensure that the backup system can take over when the main system is damaged. If the damage is serious, develop an emergency towing plan to ensure that the ship can be safely towed to the port. Record all relevant operation and status data for subsequent analysis and improvement.

[0086] Embodiment 2

[0087] As a further example, please refer to Figure 4 ,The steps of building a deep learning model through feature parameters and training the model include:

[0088] S131: inputting the characteristic parameters into the deep learning model, calculating the weight values ​​of each layer of the neural network, and outputting the predicted value of the ship collision damage;

[0089] S132: Calculate the error value between the predicted value and the historical data, and when the error value is greater than a preset error value, update the weight value of each layer of the neural network;

[0090] S133: Input the updated weight value into the deep learning model and perform iterative training until a predetermined number of iterations is reached or the calculated error value is no greater than a preset error value.

[0091] Specifically, in S131, the characteristic parameters are input into the deep learning model, the weight values ​​of each layer of the neural network are calculated, and the predicted value of the ship collision damage is output. These characteristic parameters include various characteristics that may affect the results of ship collision damage, such as hull structure strength, sailing speed, ship size, type and distribution of obstacles, route complexity, etc. The weight values ​​of the collision damage characteristic parameters in each layer of the neural network are calculated, and the collision damage prediction value is output. The deep learning model performs a series of transformations and calculations on the input data through a multi-layer neural network, and finally obtains the weight values ​​corresponding to each layer of network nodes (neurons). These weight values ​​represent the importance of each node in the network to the output prediction value.

[0092] In S132, the error value between the predicted value and the historical data is calculated. When the error value is greater than the preset error value, the weight value of each layer of the neural network is updated, and the deep learning model generates a predicted value of the collision damage based on the current weight configuration. Then, the difference between the predicted value and the actual historical data of the ship collision damage is measured by loss functions such as mean square error, cross entropy, etc., that is, the prediction error. When the error is greater than the preset error value, the weight value is updated. If the calculated error is greater than the preset threshold, it means that the prediction performance of the model has not met expectations. At this time, it is necessary to calculate the partial derivative of the error with respect to each weight through the back propagation algorithm, and use optimization algorithms such as gradient descent to update the weight value to reduce the prediction error.

[0093] In S133, the updated weight value is input into the deep learning model for iterative training until a predetermined number of iterations is reached or the calculated error value is no greater than a preset error value. Model training is an iterative process, and each round of iteration will recalculate the predicted value and error based on the new weight value, and further adjust the weight. This process will be repeated until the termination condition is met. In this training process, a maximum number of iterations is usually set to prevent overfitting. At the same time, if the error reduction is not obvious after multiple consecutive iterations or the error has dropped to a preset acceptable range, the training can also be terminated in advance.

[0094] When the training process is completed, a deep learning model that can be used to predict collision damage risk results is obtained. This model contains all the parameters and weight values ​​obtained through learning, and can be used for actual collision damage prediction tasks. The saved deep learning model can be loaded into the prediction system, which receives new ship operation data and water environment data in real time, outputs the corresponding collision damage risk results, and provides support for the safe management of ship water path planning.

[0095] Embodiment 3

[0096] As a further example, see Figure 2, before the step of obtaining historical data of ship collision damage, it also includes: obtaining real-time data of ship navigation, and determining whether there is a collision risk between the ship and the obstacle based on the real-time data. Among them, the step of determining whether there is a collision risk between the ship and the obstacle based on the real-time data of the water environment is as follows:

[0097] S210: Calculate the closest encounter distance and closest encounter time between the ship and each obstacle in the water environment using real-time data;

[0098] S230: Determine the obstacle with the greatest collision risk according to the closest approach distance and the closest approach time;

[0099] S250: Use the GJK algorithm to calculate the shortest distance between the ship and the obstacle with the greatest collision risk;

[0100] S270: Determine whether the shortest distance is less than a preset safety threshold, if so, output step S271; if not, output step S272;

[0101] S271: There is a risk of collision;

[0102] S272: There is no risk of collision.

[0103] Specifically, in this embodiment, the acquisition of real-time data of the water environment is obtained by establishing a dynamic environment model, and the existing static environment model provides a basic framework for dynamic environment modeling. The water characteristics, obstacle distribution, navigable water boundary and other information determined in the static model are still applicable in the dynamic model, and the dynamic environment modeling is expanded on the basis of the static model. By introducing the time dimension and dynamic obstacles such as other moving ships and floating objects, the static environment is transformed into a dynamic model that can reflect the actual navigation situation of the water in real time. The purpose of establishing a static environment model is to create a model that can accurately reflect the characteristics of the actual water area, so as to provide real-time water environment data in this environment, facilitate the prediction of the collision damage risk results of ships and the water route planning of ship navigation. GIS technology is used to collect geographical information of the water area, including waterways, coastlines, terrain, etc. GIS data is usually provided in the form of digital maps for generating a three-dimensional model of the water area. Sonar technology or other underwater measurement technology is used to obtain water depth data. Using three-dimensional modeling technology, the water environment and obstacles are presented in the form of a three-dimensional model.

[0104] In the application of the GJK algorithm, the dynamic environment modeling in advance can know the location, shape, size and other information of obstacles in advance, which can reduce the search range of the GJK algorithm in the search space and reduce the data preprocessing work required by the GJK algorithm before collision prediction, such as dividing obstacles into convex polygons, etc., thereby improving the efficiency of collision detection. It can also avoid misjudgment of the GJK algorithm in collision prediction, such as mistaking shadows for obstacles. The future movement trajectory of dynamic obstacles can also be predicted, which can help the GJK algorithm more accurately predict the collision probability between dynamic obstacles and ships.

[0105] See also Figure 3 , Figure 3 This is the basic application flow chart of the GJK algorithm in this process. The specific application process of the GJK algorithm in this process is:

[0106] S251: Initialization

[0107] The obstacle with the greatest risk of collision between the ship and the vehicle is selected, and the closest point of approach (CPA) and the time to close point of approach (TCPA) are calculated based on the object's speed, direction, water flow velocity and other data in the dynamic environment model. Then, a convex shape is drawn based on the size and shape of the target obstacle in the dynamic environment model, which are recorded as A and B.

[0108] Choose a starting point, usually the center of A or a known point, and set it as the vertex of the starting simple shape SS of the GJK algorithm.

[0109] S252: Create support points

[0110] For the current simple shape SS, calculate its minimum distance from shape A and shape B. This minimum distance point is called a support point, which is a point on shape A or B such that the angle between the vector from SS to this point and the boundary normal vectors of shapes A and B is the smallest.

[0111] S253: Extending Simple Shapes

[0112] Add the newly calculated support point to the simple shape SS. This is usually done by connecting the point with points already in SS, forming a polygon.

[0113] S254: Determine intersection

[0114] Check whether the simple shape SS is completely within the boundaries of shape A and shape B. If so, then shapes A and B intersect, and step S242 is repeated; if not, proceed to the next step.

[0115] S255: Calculate the shortest distance and collision point

[0116] If the simple shape SS eventually contains the starting point, then shapes A and B intersect, and the shortest distance and collision point can be calculated. The shortest distance is the distance from the starting point to the nearest point on the boundary of the simple shape SS, and the collision point is the position of this nearest point.

[0117] If the simple shape SS does not contain the starting point, then shapes A and B do not intersect, but the shortest distance between them can still be calculated, which is the distance from the starting point to the nearest point on the boundary of the simple shape SS.

[0118] If the calculated shortest distance is less than a preset safety threshold, it is determined that there is a collision risk between the ship and the obstacle with the greatest risk.

[0119] S256: Decision Analysis

[0120] The next action can be decided based on the calculation results combined with the dynamic environment model. Dynamic environment modeling can help the machine understand the characteristics of the water surface environment, including the location, shape, size, and number of obstacles, providing basic information for path planning.

[0121] (1) If shapes A and B do not intersect, the vessel automatically avoids obstacles based on the shortest distance between them, which may include changing course, adjusting speed, etc., to ensure that a sufficient safe distance is maintained from the dynamic obstacle.

[0122] (2) If the automatic obstacle avoidance function cannot be achieved, the collision damage risk outcome is predicted through a deep learning model, and the decision is made by calculating the method that minimizes the damage.

[0123] (3) If shapes A and B intersect, the damage degree of the aircraft body is analyzed by combining the deep learning model trained in advance, so as to execute the program set in advance by humans. The treatment plan for the damage degree of the aircraft body is made according to the specific navigation goal. For example, the treatment of maritime rescue missions and maritime garbage cleaning missions is different. The maritime garbage cleaning mission pays more attention to the long-term operation of the machine. When the damage degree reaches 30%, it may choose to return automatically. The rescue mission pays more attention to the completion of the mission. It may choose to return automatically when the damage degree reaches 90%. These strategies take into account the maneuverability of the ship and the current navigation environment to achieve more efficient completion of navigation goals.

[0124] Embodiment 4

[0125] As a further example, please refer to Figure 5 The steps to calculate the risk of ship collision damage include:

[0126] S141: passing and calculating real-time data in each neural network layer of the deep learning model;

[0127] S142: Output the risk results of the collision damage type and the collision damage degree according to the calculations in each layer of the neural network based on the real-time data.

[0128] By inputting real-time data of ship operation and water environment into the deep learning model, the type and severity of possible collision damage can be effectively predicted. This process depends not only on the knowledge learned by the model during the training phase, but also on the accurate processing and analysis of real-time data. The application of predicted risk results helps to improve the navigation safety of ships, reduce economic losses, and enhance the overall safety management level of the shipping industry.

[0129] Embodiment 5

[0130] See also Figure 6 , Figure 6 is a schematic diagram of a water area path planning device according to an embodiment of the present invention. Figure 1 In order to realize the water path planning method shown in FIG. 1 , this embodiment provides a water path planning device 100. Figure 1 Corresponding to the method embodiment shown, the water path planning device 100 includes:

[0131] A data acquisition unit 10 is used to acquire historical data of ship collision damage and real-time data of ship navigation;

[0132] A construction unit 20 is used to determine characteristic parameters of ship collision damage according to historical data, and to construct a deep learning model through the characteristic parameters and perform model training;

[0133] The calculation unit 30 is used to input the real-time data of the ship's navigation into the deep learning model to calculate the risk result of the ship's collision damage;

[0134] The execution unit 40 is used to plan the water path for the ship based on the risk result.

[0135] The beneficial effects of the water area path planning device 100 of the embodiment of the present invention are equivalent to the beneficial effects of the above-mentioned water area path planning method, which will not be described in detail here.

[0136] Embodiment 6

[0137] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and the program is used to implement the steps of the above-mentioned water area path planning method.

[0138] The beneficial effects of the computer-readable storage medium of the present invention are equivalent to the beneficial effects of the above-mentioned water path planning method, which will not be repeated here.

[0139] The above is only for explaining the implementation mode of the present invention and is not intended to limit the present invention. For those skilled in the art, any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention without creative work should be included in the protection scope of the present invention.

Claims

1. A water path planning method, It is characterized in that The following steps are involved: When there is a risk of collision, obtain historical data on ship collision damage; Determining characteristic parameters of ship collision damage based on the historical data; Constructing a deep learning model through the characteristic parameters and performing model training; Inputting the real-time data of ship navigation into the deep learning model to calculate the risk results of ship collision damage; The water route planning for the ship is carried out based on the risk results.

2. The water area path planning method according to claim 1, It is characterized in that The step of "obtaining historical data of ship collision damage" includes: The historical data of the ship's operation during the collision damage and the historical data of the water environment in which the ship is located are obtained, wherein the historical data of the ship's operation include the cost data of the degree of damage to the ship, the cost data of obstructions to navigation, and the cost data of the length of the path along which the damaged ship travels.

3. The water area path planning method according to claim 2, It is characterized in that The step of "determining characteristic parameters of ship collision damage according to the historical data" comprises: The cost value is calculated by a characteristic cost extraction function, and the characteristic parameters of the ship collision damage are determined according to the cost value; the characteristic cost extraction function is: F = α·L+(1-α)·C+T, α is the weight of the path length cost data, (1-α) is the weight of the obstruction cost data, T is the cost data of the degree of damage to the hull, C is the obstruction cost data, and L is the path length cost data; F represents the cost value.

4. The water area path planning method according to claim 3, It is characterized in that The step of "building a deep learning model through the characteristic parameters and performing model training" includes: Input the characteristic parameters into the deep learning model, calculate the weight values ​​of each layer of the neural network, and output the predicted value of the ship collision damage; Calculating an error value between the predicted value and the historical data, and when the error value is greater than a preset error value, updating the weight values ​​of each layer of the neural network; The updated weight value is input into the deep learning model for iterative training until a predetermined number of iterations is reached or the calculated error value is no greater than the preset error value.

5. The water area path planning method according to claim 4, It is characterized in that In the step of "updating the weight values ​​of each layer of the neural network", the weights are updated using the gradient descent method.

6. The water area path planning method according to any one of claims 1 to 5, It is characterized in that The step of "obtaining historical data of ship collision damage" also includes: obtaining real-time data of ship navigation, and determining whether there is a collision risk between the ship and the obstacle based on the real-time data.

7. The water area path planning method according to claim 6, It is characterized in that The step of "determining whether there is a collision risk between the ship and the obstacle according to the real-time data" includes: Utilizing the real-time data to calculate the closest encounter distance and closest encounter time between the ship and each obstacle in the water environment; Determine the obstacle with the greatest collision risk according to the closest approach distance and the closest approach time; Using the GJK algorithm to calculate the shortest distance between the ship and the obstacle; It is determined whether the shortest distance is less than a preset safety threshold. If so, it is determined that there is a collision risk; otherwise, there is no collision risk.

8. The water area path planning method according to any one of claims 1 to 5, It is characterized in that The step of "calculating the risk result of ship collision damage" comprises: Passing and calculating the real-time data in each neural network layer of the deep learning model; The risk results of the collision damage type and the collision damage degree are outputted based on the calculation of the real-time data in each layer of the neural network.

9. A water path planning device, It is characterized in that The device comprises: A data acquisition unit, used to acquire historical data of ship collision damage and real-time data of ship navigation; A construction unit, used to determine characteristic parameters of ship collision damage according to the historical data, and to construct a deep learning model through the characteristic parameters and perform model training; A calculation unit, used to input the real-time data of ship navigation into the deep learning model to calculate the risk result of ship collision damage; An execution unit is used to plan a water path for a ship based on the risk result.

10. A computer storage medium, It is characterized in that The computer storage medium stores a computer program for implementing the steps of the method according to any one of claims 1 to 8.

Citation Information

Cited By

  • Port infrastructure performance monitoring system based on digital twinning

    CN119309630A

  • Port infrastructure performance monitoring system based on digital twin

    WO2026130203A1