Method and device for ship collision avoidance decision-making in restricted waters based on uncertainty modeling

Abstract

Method and device for ship collision avoidance decision-making in restricted waters based on uncertainty modeling. The method includes: combining a ship motion model with a random variable representing the uncertainty of ship maneuverability to construct a motion model considering the uncertainty of ship maneuverability; constructing a ship observation model considering observation uncertainty using a Gaussian distribution; simulating the steering and speed changes of the ship using a Gaussian distribution to construct an uncertainty model for the ship's compliance with rules; capturing the motion uncertainty of other ships during navigation through a deep Gaussian process model and constructing a prediction model for the trajectories of other ships considering motion uncertainty; constructing an optimization model for ship collision avoidance behavior decision-making based on the above uncertainty models and solving the optimal collision avoidance behavior decision for the ship during navigation. The present invention comprehensively considers various uncertainty factors encountered by ships during navigation in complex and dynamic restricted waters, and establishes an optimization model for ship collision avoidance behavior decision-making with strong adaptability to effectively make ship collision avoidance decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of ship collision avoidance, and particularly to a ship collision avoidance decision-making method and device based on uncertainty modeling in restricted waters. Background Art

[0002] With the increasing global demand for maritime transportation, the navigation challenges in restricted waters have become increasingly severe, and collision accidents have become the main safety issue. Data shows that more than 40% of global ship accidents every year are caused by collisions, and nearly 80% of them are related to human decision-making errors. In restricted waters, environmental conditions such as wind speed, wind direction, visibility, and tides change frequently, seriously affecting ship maneuverability. In addition, the movements of other ships are full of uncertainties, and equipment failures, route adjustments, and human errors make accurate prediction difficult. The sensors relied on by ships (such as radar and GPS) may also lead to misjudgments due to measurement errors or information delays, further increasing the accident risk. Therefore, navigation planning and collision avoidance decision-making must fully consider these uncertainty factors to improve safety and efficiency.

[0003] Although existing collision avoidance algorithms have made certain progress, there are still the following problems: (1) Ship maneuvering is affected by environmental factors and its own parameters, and many algorithms fail to fully consider these uncertainties, affecting the reliability of decision-making. (2) Most studies rely on linear prediction of the current motion state and fail to accurately capture the complex uncertainties in ship motion, resulting in a lag in collision avoidance decision-making. (3) Many algorithms assume that ships always comply with navigation rules, but in emergency situations, ships may flexibly adjust their behaviors, and existing algorithms fail to effectively handle such situations. Summary of the Invention

[0004] The purpose of the present invention is to overcome the above-mentioned defects and problems existing in the prior art, and provide a ship collision avoidance decision-making method and device based on uncertainty modeling in restricted waters, which comprehensively considers various uncertainty factors encountered by ships when navigating in complex and dynamic restricted waters, and establishes an adaptable ship collision avoidance behavior decision-making optimization model to effectively make ship collision avoidance decisions.

[0005] To achieve the above purpose, the technical solution of the present invention is: A ship collision avoidance decision-making method based on uncertainty modeling in restricted waters, including:

[0006] Combining a ship motion model with a random variable representing the uncertainty of ship maneuverability to construct a motion model considering the uncertainty of ship maneuverability;

[0007] Adopting a Gaussian distribution to construct a ship observation model considering observation uncertainty;

[0008] Adopting a Gaussian distribution to simulate the steering and speed changes of a ship to construct an uncertainty model for ships to comply with rules;

[0009] Capture the motion uncertainty of other ships during navigation through a deep Gaussian process model, and construct a prediction model for the trajectories of other ships considering motion uncertainty;

[0010] Based on the motion model considering ship maneuverability uncertainty, the ship observation model considering observation uncertainty, the uncertainty model of ship compliance with rules, and the prediction model for the trajectories of other ships considering motion uncertainty, construct an optimization model for ship collision avoidance behavior decision-making, and solve the optimal collision avoidance behavior decision for the ship during navigation.

[0011] The optimization model for ship collision avoidance behavior decision-making is as follows:

[0012] ;

[0013] ;

[0014] ;

[0015] ;

[0016] ;

[0017] ;

[0018] In the formula, is the objective function; is the control variable at each time step; is the time step; is the minimum distance between the ship and the static obstacle; is the safety distance threshold between the ship and the obstacle; is the minimum distance between the ship and the dynamic obstacle; is the th step state of the ship; is the th step state of the ship; is the motion model considering ship maneuverability; is the mean value of the control variable of the ship at the th step; is the mean value of the control variable at the th step is and the variance is Gaussian distribution; is the variance of the control variable at the th step; is the objective function of the error between the current state and the target state; is the objective function of the control variable; is the objective function for avoiding static obstacles; Objective function for avoiding dynamic obstacles is the estimated state of the ship at the step; is the desired trajectory; is the state weight matrix; is the transpose; is the control variable weight matrix; is the minimum distance between the ship and the static obstacle at the step; is the number of dynamic obstacles; is the minimum value of the distance between the ship and the predicted positions of the obstacles at time.

[0019] The method of combining the ship motion model with random variables representing the uncertainty of ship maneuverability to construct a motion model considering the uncertainty of ship maneuverability includes:

[0020] Construct the following ship motion model:

[0021] ;

[0022] ;

[0023] In the formula, is the ship motion state at time; , , , , are respectively the vertical coordinate position, horizontal coordinate position, heading angle, speed, and yaw angular velocity of the ship at time; is the state transition matrix; and are respectively the derivatives of the ship speed and yaw angular velocity at time; is the time interval;

[0024] Define the control variable as , where and are respectively the turning angular velocity and rudder angle of the ship at time;

[0025] The relationship between the derivative of the ship yaw angular velocity and the rudder angle is expressed as:

[0026] ;

[0027] In the formula, and They are the ship's following index and the ship's turning index respectively;

[0028] The derivative of the ship's speed is:

[0029] ;

[0030] ;

[0031] ;

[0032] In the formula, is the ship's mass; is 's coefficient; is 's coefficient; is the hydrodynamic derivative; and are the propeller thrust coefficient and the drag coefficient respectively;

[0033] The control function is:

[0034] ;

[0035] Combining the control variable with the ship motion model to construct a motion model considering ship maneuverability:

[0036] ;

[0037] In the formula, is the control input matrix, indicating the influence of the control variable on the ship's state;

[0038] Simplify the motion model considering ship maneuverability to:

[0039] ;

[0040] Using random variables to represent the uncertainty of ship operations, the motion model considering the uncertainty of ship maneuverability is:

[0041] ;

[0042] In the formula, represents a Gaussian distribution with a mean of and a covariance matrix of ;

[0043] The ship observation model considering observation uncertainty is:

[0044] ;

[0045] ;

[0046] ;

[0047] In the formula, is the observed state of the ship obtained from the AIS at time is the motion state of the ship at time is the observation matrix; represents a Gaussian distribution with a mean of and a covariance matrix of

[0048] The uncertainty model for the ship to comply with the rules is:

[0049] ;

[0050] In the formula, is the control variable; and are the mean and variance of the ship's turning angular velocity respectively; and are the mean and variance of the ship's rudder angle respectively.

[0051] Capturing the motion uncertainty of other ships during navigation through a deep Gaussian process model and constructing a prediction model for the trajectory of other ships considering motion uncertainty includes:

[0052] Defining a sliding window and normalizing the ship's latitude and longitude data using the mean and variance of the ship's trajectory;

[0053] Constructing a deep Gaussian process model including a first-layer Gaussian process model and a second-layer Gaussian process model; The first-layer Gaussian process model is:

[0054] ;

[0055] ;

[0056] In the formula, is the predicted value of the ship's trajectory; is the coordinate of the ship's trajectory point ; is the first-layer Gaussian process model; is the mean function; is the kernel function, used to measure the correlation between input points; is the signal variance; is the length scale; is the coordinate of the ship's trajectory point ; ​

[0057] Given the input trajectory points and the corresponding outputs , the posterior distribution of the Gaussian process is calculated by Bayes' rule, and the posterior distribution is:

[0058] ;

[0059] wherein, is the hyperparameter; is the predicted mean; is the predicted covariance matrix;

[0060] The predicted mean is:

[0061] ;

[0062] wherein, is the covariance matrix of the input trajectory points;

[0063] The predicted covariance matrix is:

[0064] ;

[0065] wherein, is the newly input trajectory point;

[0066] The hyperparameter is determined by maximizing the marginal likelihood function, and its form is:

[0067] ;

[0068] wherein, is the number of samples of the input trajectory points;

[0069] The second-layer Gaussian process model uses the predicted output of the first-layer Gaussian process model as the input for modeling;

[0070] Calculate the uncertainty estimates of the two-layer Gaussian process model and combine the uncertainties. The combined uncertainty estimate is:

[0071] ;

[0072] wherein, is the uncertainty predicted by the first-layer Gaussian process model; is the uncertainty predicted by the second-layer Gaussian process model;

[0073] On each sliding window, the trajectory of the other ship is predicted by the deep Gaussian process model.

[0074] Solving the optimal collision avoidance behavior decision of a ship during navigation includes:

[0075] Given the initial control variables, iterative optimization is performed through the IPOPT solver to gradually approach the optimal solution of the objective function of the ship collision avoidance behavior decision optimization model;

[0076] In each iteration, the optimization direction is evaluated by calculating the gradient information of the objective function, and it is judged whether the convergence condition meets the convergence criterion;

[0077] When the change of the objective function meets the convergence criterion, the optimal control variables of the current step are output to ensure that the ship safely avoids obstacles.

[0078] A ship collision avoidance decision-making device based on uncertainty modeling, which is applied to the above-mentioned method. The device includes:

[0079] A multi-dimensional uncertainty model construction module is used to combine the ship motion model with random variables representing the uncertainty of ship maneuverability to construct a motion model considering the uncertainty of ship maneuverability; a Gaussian distribution is used to construct a ship observation model considering observation uncertainty; a Gaussian distribution is used to simulate the steering and speed changes of the ship to construct an uncertainty model for the ship to comply with the rules; a deep Gaussian process model is used to capture the motion uncertainty of other ships during navigation, and a prediction model for the trajectories of other ships considering motion uncertainty is constructed;

[0080] A ship collision avoidance behavior decision generation module is used to construct a ship collision avoidance behavior decision optimization model based on the motion model considering the uncertainty of ship maneuverability, the ship observation model considering observation uncertainty, the uncertainty model for the ship to comply with the rules, and the prediction model for the trajectories of other ships considering motion uncertainty, and solve the optimal collision avoidance behavior decision of the ship during navigation.

[0081] A ship collision avoidance decision-making device based on uncertainty modeling includes a memory and a processor;

[0082] The memory is used to store computer program code and transmit the computer program code to the processor;

[0083] The processor is used to execute the above-mentioned method according to the instructions in the computer program code.

[0084] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method is implemented.

[0085] Compared with the prior art, the beneficial effects of the present invention are:

[0086] In a ship collision avoidance decision-making method and device based on uncertainty modeling according to the present invention, a comprehensive multi-dimensional uncertainty model is constructed by quantifying the uncertainties of ship motion characteristics, surrounding environment, and the behaviors of other ships. Compared with traditional single-factor modeling methods, this model can more comprehensively and accurately reflect various uncertainty factors that may be encountered in actual navigation, improving the accuracy of the decision-making process. By combining the uncertainty model with a multi-objective optimization algorithm, the ship's maneuvering behavior can be adjusted in real time. Based on real-time navigation data and environmental feedback, the ship's heading and speed can be flexibly optimized, making the ship collision avoidance behavior more intelligent and adaptable, and significantly reducing the collision risk caused by environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 is a flowchart of a ship collision avoidance decision-making method based on uncertainty modeling according to the present invention.

[0088] Figure 2 is a schematic diagram of a ship coordinate system in an embodiment of the present invention.

[0089] Figure 3 is a schematic diagram of ship avoidance rules in an embodiment of the present invention.

[0090] Figure 4 is a flowchart of a deep Gaussian process in an embodiment of the present invention.

[0091] Figure 5 is a schematic diagram of multi-objective optimization principle in an embodiment of the present invention.

[0092] Figure 6 is a structural block diagram of a ship collision avoidance decision-making device based on uncertainty modeling according to the present invention.

[0093] Figure 7 is a structural block diagram of a ship collision avoidance decision-making device based on uncertainty modeling according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0094] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments.

[0095] See Figure 1 , a ship collision avoidance decision-making method based on uncertainty modeling, comprising:

[0096] S1. Combine the ship motion model with random variables representing the uncertainty of ship maneuverability to construct a motion model considering the uncertainty of ship maneuverability. The maneuverability of a ship is restricted by the dynamic equations. Especially in restricted waters, considering its shallow water effect, establish a ship motion model, introduce random variables to represent the uncertainty during the maneuvering process, and combine these random variables with the dynamic model to establish a ship motion model containing uncertainty, so as to more accurately predict the future navigation state of the ship.

[0097] Adopt the Gaussian distribution to construct a ship observation model considering the observation uncertainty. Shore-based monitoring equipment, on-board radar, AIS data, and inter-ship communication have certain errors, resulting in certain uncertainty in the obtained ship observation data. Use the Gaussian distribution to fuse these uncertainties with the observation model.

[0098] Use the Gaussian distribution to simulate the turning and speed changes of the ship to construct an uncertainty model for the ship's compliance with rules. In emergency situations, the ship may not fully comply with the rules. To simulate this behavior, use the Gaussian distribution to describe the turning and speed changes of the ship.

[0099] Capture the motion uncertainty of other ships during navigation through a deep Gaussian process model, and construct a prediction model for the trajectory of other ships considering the motion uncertainty. By constructing a deep Gaussian process model, capture the motion uncertainty of other ships during navigation. This model is used to capture the basic characteristics of the motion of other ships and preliminarily quantify its uncertainty, so as to provide a more accurate prediction of the trajectory of other ships.

[0100] S2. Based on the motion model considering the uncertainty of ship maneuverability, the ship observation model considering the observation uncertainty, the uncertainty model for the ship's compliance with rules, and the prediction model for the trajectory of other ships considering the motion uncertainty, construct an optimization model for ship collision avoidance behavior decision-making, and solve the optimal collision avoidance behavior decision for the ship during navigation.

[0101] This invention focuses on considering various uncertainty factors during navigation. By modeling the uncertainties in ship maneuvering, observation, rule compliance, and the motion of other ships, introduce random variables, Gaussian distribution, and deep Gaussian process respectively, establish a ship motion model, observation model, and rule compliance model containing uncertainty, especially simulate the behavior of the ship not fully complying with the rules in emergency situations. Based on these uncertainties, construct a multi-objective optimization problem to generate the optimal collision avoidance decision. The optimization objectives include the comprehensive trade-off of energy consumption, navigation economy, safety, and collision risk, so as to improve the safety of the ship during navigation in restricted waters and its ability to cope with uncertainties.

[0102] Further, constructing a motion model considering the uncertainty of ship maneuverability by combining the ship motion model with random variables representing the uncertainty of ship maneuverability includes:

[0103] Using a three-degree-of-freedom model to model the motion of the ship in three degrees of freedom: surge, sway, and yaw. The ship coordinate system is as shown in Figure 2 At any moment The state of the ship can be expressed as:

[0104] ;

[0105] Converting the continuous dynamic equation of the ship into a discrete form to construct the following ship motion model:

[0106] ;

[0107] In the formula, Is the Motion state of the ship at time; , , , , Are respectively The ordinate position, abscissa position, heading angle, speed, and yaw angular velocity of the ship at time; Is the state transition matrix; And Are respectively The derivatives of the ship speed and yaw angular velocity at time. In actual operation, the turning angular velocity Of the propeller and the rudder angle Are used to change And ; Is the time interval;

[0108] Define the control variable as , where And Are respectively The turning angular velocity and rudder angle of the ship at time;

[0109] The MMG equation of the ship is expressed as:

[0110] ;

[0111] In the formula, Is the ship mass; Is the moment of inertia about the Z-axis (in the attached figure coordinate system, pointing to the center of the earth); , , , , , , , , , are the hydrodynamic derivatives in the fixed coordinate system of the hull.

[0112] Separate the pitching motion from the rolling and yawing motions. Through Laplace transform, the relationship between the derivative of the ship's yaw angular velocity and the rudder angle is expressed as:

[0113] ;

[0114] In the formula, and are the ship's following index and the ship's turning index respectively, which can be calculated from the data of the ship's Z-shaped steering experiment.

[0115] The above relationship is called the KT maneuvering response model; considering the maneuvering uncertainty of the ship, even for the same ship, and will change under different environments. Especially in narrow waters, under the influence of the shallow water effect, the ship's resistance increases, the turning diameter is larger than that in deep water, and the ship's maneuverability is poor; at the same rudder angle, the ship's turning speed decreases.

[0116] The acceleration of the ship is controlled by the propeller thrust and the water resistance, and can be expressed as:

[0117] ;

[0118] The derivative of the ship's speed is:

[0119] ;

[0120] ;

[0121] ;

[0122] In the formula, is the ship's mass; is 's coefficient; is 's coefficient; is the hydrodynamic derivative; and are the propeller thrust coefficient and the resistance coefficient respectively;

[0123] The control function is:

[0124] ;

[0125] The control variable Combined with the ship motion model, a motion model considering ship maneuverability is constructed:

[0126] ;

[0127] In the formula, is the control input matrix, representing the influence of control variables on the ship state, in the form of a diagonal matrix, and the last two diagonal elements are equal to , and the other elements are equal to 0;

[0128] The motion model considering ship maneuverability is simplified to:

[0129] ;

[0130] Considering that there is a certain uncertainty in ship maneuvering, a random variable is used to represent the uncertainty of ship operation, then the motion model considering the uncertainty of ship maneuverability is:

[0131] ;

[0132] In the formula, represents a Gaussian distribution with a mean of and a covariance matrix of .

[0133] Furthermore, a ship observation model considering observation uncertainty is constructed through the following steps:

[0134] The position of the ship can be observed through the sensor system, and its observation model is expressed as:

[0135] ;

[0136] ;

[0137] Considering that there are certain errors in sensor devices such as AIS, the observation model of the ship is expressed as:

[0138] ;

[0139] In the formula, is the observed state of the ship obtained from AIS at time; is the motion state of the ship at time; is the observation matrix; represents a Gaussian distribution with a mean of and a covariance matrix of .

[0140] Furthermore, for inland river navigation, when two ships meet head-on, except in special circumstances, they should each turn to the right to avoid; when crossing, the ship with the other ship on its starboard side should give way to the other ship. The specific situation is as Figure 3 shown. By setting a Gaussian distribution to simulate the turning and speed changes of ships, the uncertainty of ships' compliance with the rules is reflected. The uncertainty model of ships' compliance with the rules is:

[0141] ;

[0142] In the formula, is a control variable; and are the mean and variance of the ship's turning angular velocity respectively; and are the mean and variance of the ship's rudder angle respectively.

[0143] For the right-turn situation mentioned in the above rules, can be set to 0.05 radians. Considering that ships may violate the rules in emergency situations, the variance can be set to a relatively large value. In this embodiment, it is set to 0.1 to increase the possibility of the ship turning left in an emergency.

[0144] Furthermore, the motion uncertainty of other ships during navigation is captured through a deep Gaussian process model, and a prediction model of the other ship's trajectory considering motion uncertainty is constructed. The whole process is as Figure 4 shown. The model consists of two layers of Gaussian processes. The first layer is used to capture the basic characteristics of the other ship's motion and preliminarily quantify its uncertainty; the second layer further captures more complex patterns and their corresponding uncertainties in the ship's navigation intention by performing a non-linear mapping on the output of the first layer, so as to provide a more accurate prediction of the other ship's trajectory. The specific implementation process is as follows:

[0145] Define a sliding window for online modeling, and normalize the ship's longitude and latitude data using the mean and variance of the ship's trajectory;

[0146] Construct a deep Gaussian process model including a first-layer Gaussian process model and a second-layer Gaussian process model; the first-layer Gaussian process model is:

[0147] ;

[0148] ;

[0149] In the formula, is the predicted value of the ship's trajectory; is the coordinate of the ship's trajectory point , which is a two-dimensional vector; is the first-layer Gaussian process model; is the mean function, usually assumed to be zero; is the kernel function, used to measure the correlation between input points; is the signal variance; is the length scale; is the coordinate of the ship trajectory point ;

[0150] Given the input trajectory point and the corresponding output , the posterior distribution of the Gaussian process is calculated by Bayes' rule, and the posterior distribution is:

[0151] ;

[0152] In the formula, is a matrix; are hyperparameters; is the predicted mean; is the predicted covariance matrix;

[0153] The predicted mean is:

[0154] ;

[0155] In the formula, is the covariance matrix of the input trajectory points;

[0156] The predicted covariance matrix is:

[0157] ;

[0158] In the formula, is the newly input trajectory point;

[0159] The hyperparameters (the length scale and signal variance of the kernel function) are determined by maximizing the marginal likelihood function (the purpose is to learn the most suitable hyperparameters , so as to obtain a Gaussian process model that can well fit the data), and its form is:

[0160] ;

[0161] In the formula, is the number of samples of the input trajectory points.

[0162] The second-layer Gaussian process model uses the predicted output of the first-layer Gaussian process model as input for modeling; by processing the prediction results of the first-layer Gaussian process model, especially for errors and residuals, the second-layer Gaussian process model can capture more details and complex correlations in the data, thereby further improving the prediction accuracy and outputting more accurate prediction results and uncertainty estimates.

[0163] Calculate the uncertainty estimates of the two-layer Gaussian process model and combine the uncertainties. The combined uncertainty estimate can provide a credible error range for each point of the trajectory prediction. The combined uncertainty estimate is:

[0164] ;

[0165] where, is the uncertainty predicted by the first-layer Gaussian process model; is the uncertainty predicted by the second-layer Gaussian process model;

[0166] On each sliding window, predict the trajectory of other ships through the deep Gaussian process model and perform inverse normalization. This step restores the normalized output to the original coordinate space for better understanding and application of the prediction results. At the same time, the inverse normalization process also takes into account the prediction uncertainty to ensure the effectiveness of the final result.

[0167] Furthermore, when generating ship collision avoidance behavior decisions, based on the uncertainties of ship maneuvering, observation, rule compliance, and the movements of other ships, construct a multi-objective optimization problem, as Figure 5 shown, aiming to solve the optimal collision avoidance decision of the ship during navigation. The goals of the optimization are to reduce energy consumption, improve the economy and safety of navigation, and reduce the collision risk. The specific implementation process is as follows:

[0168] Path tracking error. By calculating the error between the current state and the target state of the ship and weighted summing it, the ship tries to move along the desired path as much as possible. Specifically, the state error is added to the objective function in quadratic form:

[0169] ;

[0170] Control input energy consumption. To minimize the energy of the control variables for maneuvering the ship, a cost term of the control variables is added to the objective function, also in quadratic form:

[0171] ;

[0172] Obstacle avoidance cost. To ensure that the ship avoids static obstacles, when the distance between the ship and the obstacle is less than the safe distance, an additional penalty will be imposed, and the specific calculation is as follows:

[0173] ;

[0174] Dynamic obstacles mainly refer to other ships. The dynamic obstacle avoidance cost predicts the possible navigation positions of these ships by considering the uncertainty of navigation and calculates the minimum distance between the own ship and these predicted positions. The calculation method of this avoidance cost is similar to that of static obstacles, but the predicted positions of the obstacles at different time points need to be considered:

[0175] ;

[0176] The objective function combines four parts: path tracking, control energy consumption, static obstacle avoidance, and dynamic obstacle avoidance:

[0177] ;

[0178] Combined with the above ship dynamics, rules, and obstacle avoidance constraints, the optimization problem is formulated as:

[0179] ;

[0180] In the above formula, is the objective function; is the control variable at each time step; is the time step; is the minimum distance between the ship and the static obstacle; is the safety distance threshold between the ship and the obstacle; is the minimum distance between the ship and the dynamic obstacle; is the th step state of the ship; is the th step state of the ship; is the motion model considering the ship's maneuverability; is the mean of the control variable of the ship at the th step; is the mean of the control variable at the th step is and the variance is of the Gaussian distribution; is the variance of the control variable at the th step; is the objective function of the error between the current state and the target state; is the objective function of the control variable; is the objective function of avoiding static obstacles; Objective function for avoiding dynamic obstacles; is the estimated state of the ship at step , which is the estimated value of the ship obtained by the Kalman filter algorithm from the motion model considering the uncertainty of ship maneuverability and the ship observation model considering the observation uncertainty; is the desired trajectory; is the state weight matrix; is transpose of; is the control variable weight matrix; is the minimum distance between the ship and the static obstacle at step ; is the number of dynamic obstacles; is the minimum value of the distance between the ship and the predicted positions of the obstacles at time.

[0181] Furthermore, the solution for the optimal collision avoidance behavior decision of the ship during navigation includes:

[0182] Given the initial control variables, iterative optimization is performed through the IPOPT solver to gradually approach the optimal solution of the objective function of the ship collision avoidance behavior decision optimization model. In each iteration of the IPOPT solver, by updating the control variables, the corresponding ship state is calculated, and the optimal solution is gradually approached through the gradient descent method to ensure the optimal strategy for collision avoidance behavior.

[0183] In each iteration, the IPOPT solver evaluates the optimization direction by calculating the gradient information of the objective function (such as path tracking error, cost of avoiding obstacles), and determines whether the convergence condition meets the convergence criterion;

[0184] When the change of the objective function meets the convergence criterion, the optimal control variables of the current step are output, including specific maneuvering actions such as the direction and acceleration of the ship, to ensure that the ship safely avoids obstacles.

[0185] By comprehensively considering the uncertainties encountered by a ship when navigating in complex and dynamic restricted waters, the present invention establishes a highly adaptable model to facilitate more effective ship collision avoidance decision-making. First, through data analysis and modeling techniques, these uncertainty factors are quantified, and corresponding mathematical models are constructed. These models not only consider the motion characteristics of the ship itself but also take into account the surrounding environment and the behavior of other ships to form a comprehensive multi-dimensional uncertainty model. Then, the uncertainty model is combined with multi-objective optimization. Multiple optimization objectives are set, and through optimization algorithms, according to real-time navigation data and environmental feedback, the ship's maneuvering behavior is flexibly adjusted. The model of the present invention has strong robustness and can cope with different types of uncertainties in complex and dynamic navigation environments, especially in restricted waters, ensuring that the ship can make appropriate collision avoidance decisions and avoiding the limitations of traditional methods in the face of changing environments.

[0186] See Figure 6 , the present invention also provides a restricted waters ship collision avoidance decision-making device based on uncertainty modeling. This device is applied to the above-mentioned restricted waters ship collision avoidance decision-making method based on uncertainty modeling. The device includes:

[0187] A multi-dimensional uncertainty model construction module, which is used to combine the ship motion model with random variables representing the uncertainty of ship maneuverability to construct a motion model considering the uncertainty of ship maneuverability; use a Gaussian distribution to construct a ship observation model considering observation uncertainty; use a Gaussian distribution to simulate the steering and speed changes of the ship to construct an uncertainty model for the ship to comply with the rules; capture the motion uncertainty of other ships during navigation through a deep Gaussian process model and construct a prediction model for the trajectory of other ships considering motion uncertainty;

[0188] A ship collision avoidance behavior decision generation module, which is used to construct an optimization model for ship collision avoidance behavior decisions based on the motion model considering the uncertainty of ship maneuverability, the ship observation model considering observation uncertainty, the uncertainty model for the ship to comply with the rules, and the prediction model for the trajectory of other ships considering motion uncertainty, and solve the optimal collision avoidance behavior decision for the ship during navigation.

[0189] See Figure 7 , the present invention also provides a restricted waters ship collision avoidance decision-making device based on uncertainty modeling, including a memory and a processor;

[0190] The memory is used to store computer program code and transmit the computer program code to the processor;

[0191] The processor is used to execute the above-mentioned restricted waters ship collision avoidance decision-making method based on uncertainty modeling according to the instructions in the computer program code.

[0192] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the above-mentioned method for making collision avoidance decisions for ships in restricted waters based on uncertainty modeling.

[0193] Generally speaking, the computer instructions for implementing the method of the present invention can be carried by any combination of one or more computer-readable storage media. A non-transitory computer-readable storage medium can include any computer-readable medium except for the signals propagating temporarily themselves.

[0194] The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0195] The computer program code for performing the operations of the present invention can be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. In particular, the Python language suitable for neural network computing and platform frameworks based on TensorFlow, PyTorch, etc. can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0196] For the above-mentioned device and non-transitory computer-readable storage medium, reference can be made to the specific description of the method for making collision avoidance decisions for ships in restricted waters based on uncertainty modeling and its beneficial effects, which will not be elaborated here.

[0197] Although the embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A ship collision avoidance decision method in restricted waters based on uncertainty modeling, characterized in that: include: The ship motion model is combined with the random variables representing the uncertainty of ship maneuverability to construct a motion model that takes into account the uncertainty of ship maneuverability. Gaussian distribution is used to construct a ship observation model that takes observation uncertainty into account; Gaussian distribution is used to simulate the ship's turning and speed changes to build an uncertainty model for ship compliance with regulations; The motion uncertainty of other ships during navigation is captured through the deep Gaussian process model, and a trajectory prediction model of other ships considering the motion uncertainty is constructed; Based on the motion model considering the uncertainty of ship maneuverability, the ship observation model considering the uncertainty of observation, the uncertainty model of ship compliance with rules and the other ship trajectory prediction model considering the uncertainty of motion, the ship collision avoidance behavior decision optimization model is constructed to solve the optimal collision avoidance behavior decision of the ship during navigation. The ship collision avoidance behavior decision optimization model is: J=J track +J control +J static_obs +J dynamic_obs ; Where, J is the objective function; U is the control variable for each time step; N is the time step; h is the minimum distance between the ship and the static obstacle; D safe is the safety distance threshold between the ship and the obstacle; g is the minimum distance between the ship and the dynamic obstacle; s k is the state of the ship at step k; s k+1 is the state of the ship at step k+1; F(s k , μ k ) is the motion model considering the ship's maneuverability; μ k is the mean value of the control variable of the ship at step k; The mean value of the control variable at step k is μ k And the variance is σ k Gaussian distribution of k is the variance of the control variable at the kth step; J track is the objective function of the error between the current state and the target state; J control is the objective function of the control variable; J static_obs is the objective function of avoiding static obstacles; J dynamic_obs The objective function for avoiding dynamic obstacles; is the estimated state of the ship at step k; s ref is the expected trajectory; Q is the state weight matrix; μ k The transpose of ; R is the control variable weight matrix; is the minimum distance between the ship and the static obstacle at step k; M is the number of dynamic obstacles; is the minimum distance between the predicted position of the ship and the obstacle at time j.

2. The method for ship collision avoidance decision-making in restricted waters based on uncertainty modeling according to claim 1 is characterized in that: The ship motion model is combined with a random variable representing the uncertainty of ship maneuverability to construct a motion model that takes into account the uncertainty of ship maneuverability, including: Construct the following ship motion model: s t =[x t ,y t ,φ t ,u t ,r t ] T ; In the formula, s t is the motion state of the ship at time t; x t ,y t ,φ t 、u t 、r t are the ordinate position, abscissa position, heading angle, speed, and yaw angular velocity of the ship at time t; A is the state transfer matrix; and are the derivatives of the ship speed and yaw rate at time t respectively; Δt is the time interval; Define the control variable as μ t =[η t , δ t ] T , where η t and δ t are the turning angular velocity and rudder angle of the ship at time t respectively; The relationship between the derivative of the ship's yaw angular velocity and the rudder angle is expressed as: In the formula, T and K are the ship following index and ship maneuverability index respectively; Derivative of ship speed for: Where m is the mass of the ship; w T for The coefficient of w R for The coefficient of is the fluid mechanics derivative; K T and K R propeller thrust coefficient and drag coefficient respectively; The control function f is: The control variable μ t Combined with the ship motion model, a motion model that takes into account the ship's maneuverability is constructed: s t =As t-1 +B·f(s t-1 ,μ t ); Where B is the control input matrix, which represents the influence of the control variables on the ship state; The motion model considering the ship's maneuverability is simplified as: s t =F(s t-1 ,m t ); Using random variables to represent the uncertainty of ship operation, the motion model considering the uncertainty of ship maneuverability is: In the formula, The mean is F(S t-1 , μ t ) and the covariance matrix is ​​a Gaussian distribution of Q.

3. The method for ship collision avoidance decision-making in restricted waters based on uncertainty modeling according to claim 1 is characterized in that: The ship observation model considering observation uncertainty is: a t =Cs t ; In the formula, o t is the observed status of the ship obtained from AIS at time t; s t is the motion state of the ship at time t; C is the observation matrix; Indicates the mean value is Cs t And the covariance matrix is ​​a Gaussian distribution of R.

4. The method for ship collision avoidance decision-making in restricted waters based on uncertainty modeling according to claim 1 is characterized in that: The uncertainty model of ship compliance is: In the formula, μ is the control variable; m η and are the mean and variance of the ship's turning angular velocity, respectively; m δ and are the mean and variance of the ship’s rudder angle, respectively.

5. The method for ship collision avoidance decision-making in restricted waters based on uncertainty modeling according to claim 1 is characterized in that: The deep Gaussian process model is used to capture the motion uncertainty of other ships during navigation, and a trajectory prediction model of other ships considering the motion uncertainty is constructed, including: Define a sliding window and use the mean and variance of the ship trajectory to normalize the ship's longitude and latitude data; A deep Gaussian process model including a first-layer Gaussian process model and a second-layer Gaussian process model is constructed; the first-layer Gaussian process model is: y(x)~gp(m(x),k(x,x')); Where y(x) is the predicted value of the ship trajectory; x is the coordinate of the ship trajectory point x; gp(m(x), k(x, x′)) is the first-layer Gaussian process model; m(x) is the mean function; k(x, x′) is the kernel function used to measure the correlation between input points; is the signal variance; l is the length scale; x′ is the coordinate of the ship track point x′; Given an input trajectory point X and the corresponding output Y, the posterior distribution of the Gaussian process is calculated using the Bayesian rule, and the posterior distribution is: Where θ is a hyperparameter; γ is the predicted mean; ∑ is the predicted covariance matrix; The predicted mean γ is: γ=K(X,X)K(X,X) -1 Y; Where K(X, X) is the covariance matrix of the input trajectory points; The predicted covariance matrix ∑ is: Σ=K(X * ,X * )-K(X * ,X)K(X,X) -1 K(X,X * ); Where, X * is the newly input trajectory point; The hyperparameter θ is determined by maximizing the marginal likelihood function, which has the form: Where n is the number of samples of input trajectory points; The second-layer Gaussian process model uses the predicted output of the first-layer Gaussian process model as input for modeling; Calculate the uncertainty estimate of the two-layer Gaussian process model and merge the uncertainties. The combined uncertainty estimate is for: In the formula, σ 1 is the uncertainty of the prediction of the first-layer Gaussian process model; σ 2 The uncertainty of the prediction for the second-level Gaussian process model; On each sliding window, the trajectory of other ships is predicted by the deep Gaussian process model.

6. The method for ship collision avoidance decision-making in restricted waters based on uncertainty modeling according to claim 1 is characterized in that: The method of solving the optimal collision avoidance behavior decision of a ship during navigation includes: Given the initial control variables, the IPOPT solver is used for iterative optimization to gradually approach the optimal solution of the objective function of the ship collision avoidance behavior decision optimization model. In each iteration, the optimization direction is evaluated by calculating the gradient information of the objective function and judging whether the convergence condition meets the convergence criteria; When the change of the objective function meets the convergence criteria, the optimal control variables of the current step are output to ensure that the ship can safely avoid obstacles.

7. A ship collision avoidance decision-making device in restricted waters based on uncertainty modeling, characterized in that: The device is applied to the method described in any one of claims 1 to 6, and the device comprises: The multi-dimensional uncertainty model building module is used to combine the ship motion model with the random variables representing the uncertainty of the ship's maneuverability to build a motion model that takes into account the uncertainty of the ship's maneuverability; Gaussian distribution is used to build a ship observation model that takes into account the observation uncertainty; Gaussian distribution is used to simulate the ship's turning and speed changes to build an uncertainty model for the ship's compliance with the rules; the deep Gaussian process model is used to capture the motion uncertainty of other ships during navigation, and a trajectory prediction model for other ships that takes into account the motion uncertainty is built; The ship collision avoidance behavior decision generation module is used to construct a ship collision avoidance behavior decision optimization model based on the motion model considering the uncertainty of ship maneuverability, the ship observation model considering the uncertainty of observation, the uncertainty model of ship compliance with rules and the other ship trajectory prediction model considering the uncertainty of motion, and solve the optimal collision avoidance behavior decision of the ship during navigation.

8. A ship collision avoidance decision-making device in restricted waters based on uncertainty modeling, characterized in that: including memory and processor; The memory is used to store computer program code and transmit the computer program code to the processor; The processor is configured to execute the method according to any one of claims 1 to 6 according to the instructions in the computer program code.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images