Ship multi-target collision avoidance method based on model prediction path integration

Abstract

The invention relates to a ship multi-target collision avoidance method based on model prediction path integration, and belongs to the technical field of ship collision avoidance. The method comprises the steps that each ship determines a plurality of random sampling control sequences of the ship and a plurality of simulation tracks corresponding to the sequences based on the positions and the navigational speeds of all ships; each ship respectively calculates the weight of each simulation track of the ship based on the positions, the navigational speeds and the simulation tracks of all ships; each ship determines a current optimal simulation track of the ship based on the weight of each simulation track of the ship; and each ship optimizes the actual control sequence of the ship based on the random sampling control sequence corresponding to the current optimal simulation track of the ship. The method solves the problems that in the prior art, due to asynchronous planning and control, ship collision avoidance navigation planning is not real-time enough, and large uncertainty exists in planning and control, and safe and efficient multi-target collision avoidance of ship navigation in the busy sea area environment is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of ship collision avoidance, and particularly relates to a multi-objective ship collision avoidance method based on model predictive path integral. Background Art

[0002] Intelligent ships have the characteristics of autonomous navigation, safety and high efficiency, representing the future development of the world's ship transportation technology. Collision avoidance technology, as the core technology of intelligent ships, has also been a research hotspot in recent years. This technology refers to that after a ship senses the surrounding sea area environment, it makes an autonomous collision avoidance decision and directly delivers the decision to the control system for execution, and then autonomously returns to the original route to continue sailing after "passing and clearing". At present, the research on ship collision avoidance methods mostly focuses on the design of collision avoidance methods in open waters, and there is less research on multi-objective ship collision avoidance methods in narrow and busy water area environments.

[0003] In the existing multi-objective collision avoidance technology, the navigation path is mostly planned at the front end, and then the underlying controller realizes the tracking control according to the planned path. The limitation caused thereby is that the planning and control are not carried out synchronously, and both the planning effect and the tracking control effect will affect the real-time performance of ship collision avoidance navigation. The coordination of planning and control needs to be adjusted manually according to the actual scenario, and the uncertainty of their planning and control effects increases; moreover, the trajectory planned at the front end may have physically infeasible results, and it is difficult for the controller to track this trajectory. Summary of the Invention

[0004] In view of the above analysis, the present invention aims to disclose a multi-objective ship collision avoidance method based on model predictive path integral, to solve the problems in the prior art that the ship collision avoidance navigation planning is not real-time enough and there are large uncertainties in planning and control due to the asynchronous progress of planning and control, and to realize multi-objective collision avoidance for ships sailing safely and efficiently in a busy sea area environment.

[0005] A multi-objective ship collision avoidance method based on model predictive path integral provided by the present invention specifically includes the following steps:

[0006] Each ship respectively determines multiple random sampling control sequences of itself and multiple simulated trajectories corresponding to each sequence based on the positions and speeds of all ships; all ships include each ship and the target collision avoidance ships corresponding to each ship;

[0007] Each ship respectively calculates the weights of its own simulated trajectories based on the positions, speeds and simulated trajectories of all ships;

[0008] Each ship respectively determines its current optimal simulated trajectory based on the weights of its own simulated trajectories;

[0009] Each ship respectively optimizes its actual control sequence based on the random sampling control sequence corresponding to its current optimal simulated trajectory.

[0010] Further, each of the vessels calculates the weights of its respective simulated trajectories based on the positions, speeds, and simulated trajectories of all the vessels, including:

[0011] Each of the vessels calculates the risk improvement cost of its respective simulated trajectories based on the positions, speeds, and simulated trajectories of all the vessels.

[0012] Each of the vessels calculates the safety cost, distance cost, and robustness cost corresponding to its respective simulated trajectories based on the positions, speeds, and its respective simulated trajectories of all the vessels.

[0013] Each of the vessels calculates the weights of its respective simulated trajectories based on the safety cost, distance cost, robustness cost, and risk improvement cost of its respective simulated trajectories.

[0014] Further, the calculation method of the risk improvement cost of each simulated trajectory includes:

[0015] Each of the vessels calculates the risk improvement value of its respective simulated trajectories based on the simulated trajectories of all the vessels.

[0016] Each of the vessels calculates the time urgency value based on the positions, speeds, and its respective simulated trajectories of all the vessels.

[0017] Each of the vessels calculates the risk improvement cost of its respective simulated trajectories based on the risk improvement value and the time urgency value.

[0018] Further, the calculation method of the risk improvement value of the simulated trajectory is:

[0019]

[0020] Wherein, represents the risk improvement value of the simulated trajectory X i of vessel i, X′ i represents the current trajectory of vessel i; X′ j,u represents the u-th simulated trajectory of the target collision avoidance vessel j of vessel i; M is the number of simulated trajectories of vessel j; N is the total number of target collision avoidance vessels of vessel i; frisk(X′ i , X′ j,u ) represents the vessel collision risk assessment value of the current trajectory X′ i and X′ j,u of vessel i; frisk(X i , X′ j ) represents the vessel collision risk assessment value of the simulated trajectory X i , and X′ j,u of vessel i.

[0021] Furthermore, the time urgency value is calculated as follows:

[0022]

[0023] in, Simulate trajectory X for ship i i The time urgency value of Indicates that in the simulation trajectory X i In this case, the minimum remaining time for the target collision avoidance ship to enter the ship's domain; β is the time threshold; α is the parameter for adjusting the exponential growth rate.

[0024] Furthermore, the calculation method of the risk improvement cost of the simulation trajectory is:

[0025]

[0026] A=max(Rimp(X i )), X i ∈{CA ni};

[0027] B=min(Rimp(X i )), X i ∈{CA ni};

[0028]

[0029] Among them, f vimp Improve the cost of risk; X i is the simulated trajectory of ship i, {CA ni} is the set of simulated trajectories of ship i, n is the number of target collision avoidance ships corresponding to ship i, Rimp(X i ) is the simulated trajectory X of ship i i The total risk value.

[0030] Furthermore, the calculation method of the safety cost is:

[0031] f risk =max{f j (DDV)|j=1,2,3…N};

[0032] Among them, f risk is the safety cost of the simulated trajectory, f j (DDV) is the ship collision risk assessment value between the target collision avoidance ship j and the current ship under the simulated trajectory; N is the number of target collision avoidance ships.

[0033] Furthermore, the calculation method of the distance cost is:

[0034]

[0035] Among them, f loss is the path cost of the simulated trajectory, V is the sailing speed, t r is the course recovery time, Δc is the course change value, R s is the turning radius when changing the course, and rad(Δc) represents converting the course change value into radians.

[0036] Furthermore, the calculation method of the robustness cost is as follows:

[0037] f rob = max{f j (PDV)|j = 1, 2, 3…N};

[0038] Among them, f rob is the robustness cost of the simulated trajectory, f j (PD) is the probability value of the target collision avoidance ship j colliding with the current ship in the ship conflict domain under this simulated trajectory; N is the number of target collision avoidance ships.

[0039] Furthermore, the calculation method of the weight of each simulated trajectory is as follows:

[0040]

[0041] S(τ) = α1f risk + α2f rob + α3f loss + α4f vimp ;

[0042] Among them, w(τ) is the weight of the simulated trajectory τ, S(τ) is the comprehensive cost of the simulated trajectory τ, λ is a positive scaling parameter, f risk , f rob , f loss , f vimp are respectively the safety cost, path cost, robustness cost, and risk improvement cost of the simulated trajectory τ, and α1, α2, α3, α4 are the corresponding weight coefficients.

[0043] The present invention can at least achieve one of the following beneficial effects:

[0044] By adopting the path integral method, the actual control sequence of the ship is optimized based on the random sampling control sequence corresponding to the current optimal simulation trajectory of the ship itself. The first control input in the optimized actual control sequence is used as the control command at the current moment to drive the system towards the desired state. Then, at the next time step, the latest state of the system is used as the new initial state, and all collision avoidance tasks are iteratively completed. The random sampling control sequence covers the uncertainties of the planning and control effects, and naturally adapts to noise and model errors, making it suitable for solving the planning and control problems in complex dynamic environments.

[0045] By randomly sampling the control sequence and observing the response of the ship system, the problem in the prior art that the planned route is uncontrollable due to sampling in the state space is solved, and the real-time performance of ship trajectory planning is improved.

[0046] By generating alternative simulation trajectories through model forward simulation and based on the heading change constraint and the collision avoidance action time constraint, the problem in the prior art that the planned trajectory may be physically infeasible is solved.

[0047] Other features and advantages of the present invention will be described in the following specification, and some advantages can be made obvious from the specification, or understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained from the content specifically pointed out in the specification, claims, and drawings. Brief Description of the Drawings

[0048] The drawings are only for the purpose of showing specific embodiments, and are not considered as a limitation of the present invention. Throughout the drawings, the same reference signs represent the same components.

[0049] Figure 1 It is a flowchart of the method of the present invention. Detailed Embodiments

[0050] The following will specifically describe the preferred embodiments of the present invention in conjunction with the drawings. The drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0051] An embodiment of the present invention discloses a multi-objective collision avoidance method for ships based on model predictive path integral, which specifically includes steps S01 - S04.

[0052] It should be noted that in the present invention, information such as the positions, speeds, and trajectories of the encountering ships (each ship and the target collision avoidance ship corresponding to each ship) can be exchanged in real time through AIS (Automatic Identification System) or other communication methods.

[0053] Step S01: Each ship determines multiple random sampling control sequences of itself and multiple simulated trajectories corresponding to each of the sequences based on the positions and speeds of all ships; all ships include each ship and the target collision avoidance ships corresponding to each ship. Step S01 includes S011 - S012.

[0054] S011: Each ship generates multiple alternative random sampling control sequences of itself and multiple alternative simulated trajectories corresponding to each of the alternative random sampling control sequences based on the positions and speeds of all ships.

[0055] Specifically, each ship constructs its own MPPI controller. The MPPI controller generates an initial sampling control sequence based on the control distribution p, and the initial sampling control sequence represents the control actions that the system may take without any optimization or adjustment, such as controlling the rotational speed n of the ship's propeller p and the corresponding control actions of the rudder angle. Among them, the control distribution p is the original probability distribution of the system control input without external information or prior knowledge.

[0056] Furthermore, multiple sampling control sequences are obtained through iteration based on the initial sampling control sequence, and the iteration process is expressed as:

[0057]

[0058] where, v i,t is the control sequence of the i-th sampling, is the optimal control sequence at the previous moment, ∈ i,t-1 is the perturbation sampled from the Gaussian distribution at the previous moment, that is Σ is the covariance matrix, which determines the magnitude and direction of the perturbation.

[0059] Furthermore, the multiple random sampling control sequences are composed of the initial sampling control sequence and the multiple sampling control sequences obtained through iteration.

[0060] Furthermore, based on the multiple random sampling control sequences, multiple complete simulated trajectories are generated by iterating the ship's maneuvering dynamic model. Specifically, the state variables defining the ship's dynamic behavior are:

[0061]

[0062] where, x and y are the position coordinates of the ship, is the heading angle of the ship, u and v are the speeds of the ship along the x and y axes of the hull respectively, and r is the yaw angular velocity of the ship.

[0063] Furthermore, the state function of the ship is defined based on the state variables of the ship as:

[0064]

[0065] Among them, u is the control variable, that is, the random sampling control sequence, including the rudder angle δ and the propeller speed n p etc., and the state equation can be further expressed as:

[0066]

[0067] Among them, are the position changes of the ship on the x and y axes respectively, is the change in the course angle of the ship, are the speed changes of the ship on the x and y axes and the change in the turning speed respectively, m xx and m yy are the added masses of the ship in the x and y axis directions in the horizontal plane respectively, I z is the moment of inertia of the ship about the vertical axis, X H 、X R 、X P are the longitudinal forces generated by the hull, rudder and propeller respectively, Y H 、Y R 、Y P are the lateral forces generated by the hull, rudder and propeller respectively, N H 、N R 、N P are the moments of the hydrodynamic force, resistance and propulsive force about the vertical axis respectively.

[0068] Furthermore, the ship's maneuvering dynamic model obtains multiple alternative simulated trajectories corresponding to the multiple alternative random sampling control sequences based on the state function of the ship. It should be noted that when calculating and generating multiple alternative simulated trajectories, the ship's maneuvering dynamic model can use an accurate mathematical model of dynamics, a model approximated by a neural network, or a simulation model run by simulation software.

[0069] Step S012: Based on the multiple alternative random sampling control sequences and each of the alternative simulated trajectories, determine multiple random sampling control sequences of each ship itself and the corresponding multiple simulated trajectories.

[0070] Specifically, a target universal set is formed based on the multiple alternative random sampling control sequences and each of the alternative simulated trajectories.

[0071] In the target universal set, select alternative random sampling control sequences that meet the course change constraint and the corresponding alternative simulated trajectories to form a first target subset.

[0072] The course change constraint is:

[0073]

[0074] The meaning of the course change constraint is that according to the International Regulations for Preventing Collisions at Sea, when ships meet, both ships avoid collision by turning to the starboard side, and the turning angle of the ship is generally less than 90 degrees.

[0075] Furthermore, in the first target subset, alternative simulated trajectories that meet the collision avoidance action time constraint and the corresponding alternative random sampling control sequences are selected to form a second target subset.

[0076] The collision avoidance action time constraint is expressed as:

[0077] 0 < t a < minTDV;

[0078] where, t a is the time for the current ship to take collision avoidance action, that is, the time for the current ship to complete the current simulated trajectory, and TDV is the remaining time for the target collision avoidance ship to enter the domain of the current ship, which is obtained during simulation by the ship's maneuvering dynamic model.

[0079] Furthermore, the second target subset is the multiple random sampling control sequences and the corresponding multiple simulated trajectories of each ship itself determined in step S01.

[0080] Step S02: Each ship calculates the weight of its own simulated trajectories based on the positions, speeds, and simulated trajectories of all ships. Specifically, it includes S021 - S023.

[0081] S021: Each ship calculates the risk improvement cost of its own simulated trajectories based on the positions, speeds, and simulated trajectories of all ships;

[0082] S022: Each ship calculates the safety cost, distance cost, and robustness cost corresponding to its own simulated trajectories based on the positions, speeds of all ships, and its own simulated trajectories;

[0083] S023: Each ship calculates the weight of its own simulated trajectories based on the safety cost, distance cost, robustness cost, and risk improvement cost of its own simulated trajectories.

[0084] Specifically, in S021, the calculation method of the risk improvement cost of each simulated trajectory includes:

[0085] Each ship calculates the risk improvement value of its own simulated trajectories based on the simulated trajectories of all ships;

[0086] Each ship calculates the time urgency value based on the positions, speeds of all ships, and its own simulated trajectories;

[0087] Each of the vessels calculates the risk improvement cost of its respective simulated trajectories based on the risk improvement value and the time urgency value.

[0088] Furthermore, the calculation method for the risk improvement value of each simulated trajectory is as follows:

[0089]

[0090] Wherein, represents the risk improvement value of the simulated trajectory X of vessel i, X′ i represents the current trajectory of vessel i; X′ i represents the u-th simulated trajectory of the target collision avoidance vessel j of vessel i; M is the number of simulated trajectories of vessel j; N is the total number of target collision avoidance vessels of vessel i; frisk(X′ j,u , X′ i ) represents the vessel collision risk assessment value of the current trajectory X′ j,u and X′ i ; frisk(X j,u , X′ i ) represents the vessel collision risk assessment value of the simulated trajectory X j , and X′ i , and X′ j,u .

[0091] Furthermore, the calculation method for the time urgency value of each simulated trajectory is as follows:

[0092]

[0093] Wherein, is the time urgency value of the simulated trajectory X of vessel i i ; represents the minimum value of the remaining time for the target collision avoidance vessel to enter the vessel domain of this vessel in the case of the simulated trajectory X i ; β is the time threshold, preferably set to 12 minutes; α is the parameter for adjusting the exponential growth rate, preferably set to 6 / ln2.

[0094] Furthermore, the calculation method for the risk improvement cost of each simulated trajectory is as follows:

[0095]

[0096] A = max(Rimp(X i ))), X i ∈{CA ni};

[0097] B = min(Rimp(X i ))), X i ∈{CA ni};

[0098]

[0099] Among them, f vimp is the risk improvement cost; X i is the simulated trajectory of ship i, {CA ni} is the set of simulated trajectories of ship i, n is the number of target collision avoidance ships corresponding to ship i, and Rimp(X i ) is the total risk value of the simulated trajectory X i of ship i.

[0100] Specifically, in S022, the calculation method of the safety cost is as follows:

[0101] f risk = max{f j (DDV)|j = 1, 2, 3…N};

[0102] Among them, f risk is the safety cost of the simulated trajectory, and f j (DDV) is the ship collision risk assessment value between the target collision avoidance ship j and the current ship under this simulated trajectory; N is the number of target collision avoidance ships.

[0103] The following is the derivation of the calculation method of f j (DDV) (including steps ①②③):

[0104] ① Parse and express the positions and relevant motion parameters of each ship as follows:

[0105] v x = vsin(c)

[0106] v y = vcos(c)

[0107] P1(t) = (x1 + v x1 t, y1 + v y1 t)

[0108] P2(t) = (x2 + v x2 t, y2 + v y2 t)

[0109] Among them, c is the ship's course angle, v is the ship's speed, v x and v y are the ship's speeds relative to the earth coordinate system, (x1, y1) and (x2, y2) respectively represent the initial position coordinates of the target collision avoidance ship and the current ship, and P1 and P2 are the position coordinates of the two ships at time t.

[0110] ② Based on the analysis in step ①, determine the minimum encounter distance between the two ships in any time period;

[0111] Specifically, the calculation formula for the distance D(t) between two ships at time t is:

[0112]

[0113] where A = (v x1 - v x2 ) 2 + (v y1 - v y2 ) 2 ;

[0114] B = 2[(x1 - x2)(v x1 - v x2 ) + (y1 - y2)(v y1 - v y2 )];

[0115] C = (x1 - x2) 2 + (y1 - y2) 2 .

[0116] Furthermore, the minimum encounter distance CPA between two ships in any time period [0, t1] can be calculated by the following formula:

[0117]

[0118] ③ Calculate the ship collision risk assessment value based on the minimum encounter distance between two ships in any time period.

[0119] Specifically, f j (DDV) = 1 - CPA j,min / R d ;

[0120] where R d is the maneuvering radius of the current ship; CPA j,min is the minimum encounter distance between the target collision avoidance ship j and the current ship during the collision avoidance process; CPA j,min = min{CPA j (t n , t n+1 )|n = 0, 1, 2, 3, 4…}, t n and t n+1 represent the time series.

[0121] Specifically, in S022, the calculation method of the distance cost is:

[0122]

[0123] where f loss is the distance cost of the simulated trajectory, V is the speed, tr is the course recovery time, Δc is the course change value, and R s is the turning radius when changing the course, and rad(Δc) represents converting the course change value to radians.

[0124] In S022, the calculation method of the robustness cost is as follows:

[0125] f rob = max{f j (PDV)|j = 1, 2, 3…N};

[0126] where f rob is the robustness cost of the simulated trajectory, and f j (PDV) is the probability value of the target collision avoidance ship j colliding with the current ship in the ship conflict domain under this simulated trajectory; N is the number of target collision avoidance ships.

[0127] It should be noted that the probability value f j (PDV) of the collision takes into account the uncertainty of ship movement, such as the limitations of ship maneuverability, the influence of environmental factors, and other unexpected situations, providing a more comprehensive risk assessment, recorded as f j (PDV) = PDV(t), and the calculation method is as follows:

[0128] PDV(t) = ∫ λ∈D N2(λ; λ ot (t), ∑ ot (t))dλ

[0129] where N2(λ; λ ot (t), ∑ ot (t)) is the two-dimensional Gaussian probability density function, λ ot (t) is the mean value of the distance between ships, ∑ ot (t) is the covariance matrix of the distance, and D is the conflict domain of the current ship.

[0130] Through coordinate transformation, the above two-dimensional Gaussian PDV(t) is decomposed into two independent one-dimensional Gaussians, expressed as:

[0131]

[0132] where ψ is the one-dimensional Gaussian cumulative distribution function, Δξ, Δη are the transformed abscissa and ordinate, and δξ(t), δη(t) are the boundaries of the extended ship conflict domain.

[0133] Furthermore, the boundaries of the extended ship conflict domain are determined through the following steps:

[0134]

[0135] Wherein, M is the error covariance matrix, and L is the lower triangular matrix obtained by Cholesky decomposition of M;

[0136] T = RL -1 ;

[0137] Wherein, R is the orthogonal rotation transformation matrix, and T is the orthogonal rotation transformation matrix of the target ship;

[0138] W = T -1 ;

[0139] Wherein, W is the inverse matrix of the transformation matrix;

[0140]

[0141] Wherein, a and c are the diagonal elements of matrix W T W, that is, the variances of the ship position prediction error in two main axis directions, representing the position uncertainties of the ship along the course and roll directions; b is the non-diagonal element representing matrix W T W, representing the correlation between errors;

[0142]

[0143] Wherein, R d is the radius of the ship domain.

[0144] Specifically, in S023, the calculation method of the weights of the simulated trajectories is as follows:

[0145]

[0146] S(τ) = α1f risk +α2f rob +α3f loss +α4f vimp ;

[0147] Wherein, w(τ) is the weight of the simulated trajectory τ, S(τ) is the comprehensive cost of the simulated trajectory τ, λ is a positive scaling parameter, f risk 、f rob 、f loss 、f vimpThey are the safety cost, path cost, robustness cost, and risk improvement cost of the simulated trajectory τ respectively. α1, α2, α3, and α4 are the corresponding weight coefficients respectively, which are used to adjust the influence of each cost on the comprehensive cost. It should be noted that the value of λ is used to adjust the influence degree of the cost function on the weights and is adjusted based on the actual task requirements. When λ is small, the weight distribution will be more concentrated. This means that the control sequence with better performance will obtain higher weights, and the algorithm tends to select the control sequence with the best performance, which may lead to local optimality of the result; while when λ is large, the weight distribution is more dispersed, the weight distribution will be smoother, and the algorithm will tend to explore more control sequences, increasing the possibility of finding the global optimal solution.

[0148] Step S03: Each of the ships determines its current optimal simulated trajectory based on the weights of its respective simulated trajectories.

[0149] Specifically, the simulated trajectory with the largest weight is the current optimal simulated trajectory.

[0150] Step S04: Each of the ships optimizes its actual control sequence based on the random sampling control sequence corresponding to its current optimal simulated trajectory.

[0151] Specifically, the path integral method of minimizing relative entropy is used to optimize the actual control sequence of the ship based on the random sampling control sequence corresponding to its current optimal simulated trajectory.

[0152] The calculation formula is:

[0153]

[0154] Among them, represents the optimized control sequence at time t; η is the normalization coefficient; V is the control action sequence corresponding to the optimal simulated trajectory; S(V) represents the comprehensive cost of the optimal simulated trajectory; p(V) is the probability density function of the basic distribution, which is used to represent the control cost of the random control input; v t is the control input variable, that is, a single control action in the control sequence; Σ is the covariance matrix, which is the same as Σ in step S11; m is the number of the control input vectors v t .

[0155] Specifically, in practical applications, the first control input in the actual control sequence optimized in step S04 is used as the control command at the current moment to drive the system towards the desired state. Then, at the next time step, the latest state of the system is used as the new initial state, and the above steps S01 - S04 are iterated until all collision avoidance tasks are completed.

[0156] A ship multi-objective collision avoidance method based on model predictive path integral disclosed in this embodiment optimizes its actual control sequence based on the random sampling control sequence corresponding to the current optimal simulation trajectory of the ship itself by adopting the path integral method, takes the first control input in the optimized actual control sequence as the control command at the current moment, promotes the system to develop towards the desired state, and then at the next time step, uses the latest state of the system as the new initial state to iteratively complete all collision avoidance tasks. The uncertainty of the planning and control effects is covered by the random sampling control sequence, and it naturally adapts to noise and model errors, which is suitable for solving the planning and control problems in complex dynamic environments.

[0157] By randomly sampling the control sequence and observing the response of the ship system, the problem that the planned route is uncontrollable due to sampling in the state space in the prior art is solved, and the real-time performance of ship trajectory planning is improved. By generating alternative simulation trajectories through model forward simulation and based on the heading change constraint and the collision avoidance action time constraint, the problem that the planned trajectory may be physically infeasible in the prior art is solved.

[0158] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A multi-objective collision avoidance method for ships based on model predictive path integral, characterized in that It includes the following steps: Each ship respectively determines multiple random sampling control sequences of itself and multiple simulated trajectories corresponding to each sequence based on the positions and speeds of all ships; all ships include each ship and the target collision avoidance ship corresponding to each ship; Each of the ships respectively calculates the weights of its own simulated trajectories based on the positions, speeds and simulated trajectories of all ships; Each of the ships respectively determines its current optimal simulated trajectory based on the weights of its own simulated trajectories; Each of the ships respectively optimizes its actual control sequence based on the random sampling control sequence corresponding to its current optimal simulated trajectory.

2. The multi-objective collision avoidance method for ships according to claim 1, wherein Each of the ships respectively calculating the weights of its own simulated trajectories based on the positions, speeds and simulated trajectories of all the ships includes: Each of the ships respectively calculates the risk improvement cost of its own simulated trajectories based on the positions, speeds and simulated trajectories of all ships; Each of the ships respectively calculates the safety cost, distance cost and robustness cost corresponding to its own simulated trajectories based on the positions, speeds of all ships and its own simulated trajectories; Each of the ships respectively calculates the weights of its own simulated trajectories based on the safety cost, distance cost, robustness cost and risk improvement cost of its own simulated trajectories.

3. The multi-objective collision avoidance method for ships according to claim 2, characterized in that, The calculation method of the risk improvement cost of each simulated trajectory includes: Each of the ships respectively calculates the risk improvement value of its own simulated trajectories based on the simulated trajectories of all ships; Each of the ships respectively calculates the time urgency value based on the positions, speeds of all the ships and its own simulated trajectories; Each of the ships calculates the risk improvement cost of its own simulated trajectories based on the risk improvement value and the time urgency value.

4. The multi-objective collision avoidance method for ships according to claim 3, characterized in that The calculation method of the risk improvement value of the simulated trajectory is: Among them, represents the risk improvement value of the simulated trajectory X of ship i, X′ i ; X′ i represents the current trajectory of ship i; X′ j,u represents the u-th simulated trajectory of the target collision avoidance ship j of ship i; M is the number of simulated trajectories of ship j; N is the total number of target collision avoidance ships of ship i; frisk(X′ i , X′ j,u ) represents the ship collision risk assessment value of the current trajectory X′ of ship i i and X′ j,u ; frisk(X i , X′ j ) represents the ship collision risk assessment value of the simulated trajectory X i , and X′ j,u .

5. The multi-objective collision avoidance method for ships according to claim 4, wherein The calculation method of the time urgency value is: Among them, is the time urgency value of the simulated trajectory X of ship i i ; represents the minimum value of the remaining time for the target collision avoidance ship to enter the ship domain of this ship in the case of the simulated trajectory X i ; β is the time threshold; α is the parameter for adjusting the exponential growth rate.

6. The multi-objective collision avoidance method for ships according to claim 5, wherein The calculation method of the risk improvement cost of the simulated trajectory is: Among them, f vimp is the risk improvement cost; X i is the simulated trajectory of ship i, {CA ni} is the set of simulated trajectories of ship i, n is the number of target collision avoidance ships corresponding to ship i, and Rimp(X i ) is the total risk value of the simulated trajectory X i of ship i.

7. The multi-objective collision avoidance method for ships according to claim 2, wherein, The calculation method of the safety cost is: f risk = max{f j (DDV)| j = 1, 2, 3…N}; where, f risk is the safety cost of the simulated trajectory, and f j (DDV) is the ship collision risk assessment value between the target collision avoidance ship j and the current ship under this simulated trajectory; N is the number of target collision avoidance ships.

8. The multi-objective collision avoidance method for ships according to claim 2, wherein, The calculation method of the distance cost is: Among them, f loss is the path cost of the simulated trajectory, V is the sailing speed, t r is the course recovery time, Δc is the course change value, R s is the turning radius when changing the course, and rad(Δc) represents converting the course change value into radians.

9. The multi-objective collision avoidance method for ships according to claim 2, characterized in that, The calculation method of the robustness cost is: f rob = max{f j (PDV)| j = 1, 2, 3…N}; Among them, f rob is the robustness cost of the simulated trajectory, f j (PDV) is the probability value that the target collision avoidance ship j collides with the current ship in the ship conflict domain under this simulated trajectory; N is the number of target collision avoidance ships.

10. The multi-objective collision avoidance method for ships according to any one of claims 2-9, characterized in that, The calculation method of the weights of the simulated trajectories is: S(τ) = α1f risk + α2f rob + α3f loss + α4f vimp ; Among them, w(τ) is the weight of the simulated trajectory τ, S(τ) is the comprehensive cost of the simulated trajectory τ, λ is a positive scaling parameter, f risk , f rob , f loss , f vimp are respectively the safety cost, the path cost, the robustness cost, and the risk improvement cost of the simulated trajectory τ, and α1, α2, α3, and α4 are the corresponding weight coefficients respectively.

Images