Rotary drum sailboat event triggering optimization control method oriented to marine ranch cruising task

Abstract

The invention discloses a marine ranch cruising task-oriented drum sailboat event triggering optimization control method. The method comprises the following steps of: constructing a three-degree-of-freedom nonlinear mathematical model of a drum sailboat; generating a virtual reference path based on the waypoints by using LVS so as to obtain a guidance law and a position error; improving the guidance law by using a limited boundary circle rule to obtain an improved ILVS guidance law, and further obtaining a heading error; designing a virtual control law based on the actor-commentator neural network; a dynamic surface control technology is introduced to filter the virtual control law and define a dynamic error, and a robust neural damping technology is used to carry out robustness processing on a nonlinear term; designing intermediate control input by combining the actor-commentator neural network, the dynamic error subjected to robust processing and the derivative of the filtered virtual control law; designing an integral dynamic event triggering mechanism, and designing a ship path tracking controller by combining an adaptive compensation technology and intermediate control input; according to the invention, the communication load of a guidance system can be effectively reduced, the practicability of the controller in a dynamic marine environment is effectively improved, and the cruise precision of a marine ranch is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent ship control, and particularly to an event-triggered optimization control method for a rotor-assisted sailboat for ocean ranch cruising missions. Background Art

[0002] In recent years, fuel consumption and marine pollution have become increasingly prominent problems in the shipping industry. Traditional ocean ranch cruising mostly relies on fuel-powered ships, which have problems such as high energy consumption and serious environmental noise pollution. As an innovative sail-assisted ship, rotor-assisted vehicles (RAVs) can effectively address this cruising problem with their unique advantages such as simple structure, convenient installation, and efficient propulsion. In addition to being equipped with conventional propellers and steering gear equipment, rotor-assisted sailboats are also equipped with Flettner rotor equipment as auxiliary propulsion power, but essentially still belong to the category of underactuated ships. Therefore, the path tracking guidance and control strategies for conventional unmanned ships are also applicable to rotor-assisted sailboats. Currently, there are still the following three defects in the autonomous cruising process of rotor-assisted sailboats in ocean ranches:

[0003] 1) Most of the guidance signals of the ship are calculated continuously and directly transmitted to the control system in real time. However, in the complex and changeable ocean ranch environment, this transmission method has obvious drawbacks. When the signal amplitude is too large or the transmission frequency is too high, it is easy to cause input saturation problems and generate a large amount of communication load.

[0004] 2) When the instantaneous fluctuations of the marine environment (such as wind waves and ocean currents) are small but there are continuous low-frequency disturbances, the traditional triggering mechanism may delay signal updates because the preset triggering threshold is not reached, resulting in a lag in course correction and even a large deviation from the cruising route, seriously threatening the safety of ocean ranch cruising.

[0005] 3) Most of the existing reinforcement learning methods are designed based on the negative gradient descent principle of the Bellman residual square. In this design mode, the constructed Actor-Critic update rule is extremely complex, resulting in an increase in controller processing delay and thus reducing the emergency response ability of the ship. Summary of the Invention

[0006] The present invention provides an event-triggered optimization control method for a rotor-assisted sailboat for ocean ranch cruising missions to overcome the technical problems in the prior art that when performing path tracking of a rotor-assisted sailboat, the controller design is complex, the signal update is delayed, resulting in channel occupancy and actuator wear, and the guidance reference signal generated by the controller causes input saturation and communication redundancy, reducing the cruising accuracy of the rotor-assisted sailboat.

[0007] To achieve the above object, the technical solution of the present invention is:

[0008] An event-triggered optimization control method for a rotary sailboat facing ocean ranch cruise missions, including:

[0009] S1: Construct a 3-degree-of-freedom nonlinear mathematical model of the rotary sailboat as the subsequent control object;

[0010] S2: Set waypoints, use LVS to generate a virtual reference path based on the waypoints, obtain the guidance law of the rotary sailboat and the position error of the rotary sailboat to LVS according to the virtual reference path; improve the guidance law of the actual rotary sailboat using the finite boundary circle rule to obtain an improved ILVS guidance law, and obtain the position error and heading error of the rotary sailboat to LVS based on the improved ILVS guidance law;

[0011] S3: Design virtual control laws for position error and heading error respectively based on the actor-critic neural network;

[0012] S4: Introduce the dynamic surface control technology to filter the virtual control laws for position error and heading error respectively, define the dynamic error based on the filtered virtual control laws, and use the robust neural damping technology to robustify the nonlinear terms in the dynamic error to obtain the robustified dynamic error;

[0013] S5: Combine the actor-critic neural network, the robustified dynamic error, and the derivative of the filtered virtual control law to design the intermediate control inputs in the displacement velocity and displacement direction;

[0014] S6: Design an integral dynamic event-triggering mechanism to judge whether the change error of the intermediate control input satisfies the integral dynamic event-triggering mechanism. If it satisfies, obtain the intermediate control input that updates the next triggering time. If it does not satisfy, keep the intermediate control input updated at the previous triggering time until a new intermediate control input is obtained at the next triggering time, and then judge whether the new intermediate control input satisfies the integral dynamic event-triggering mechanism;

[0015] S7: Combine the adaptive compensation technology to design a ship path tracking controller based on the updated intermediate control input. The ship path tracking controller is used to perform path tracking control on the rotary sailboat to realize the autonomous cruise mission of the rotary sailboat in the ocean ranch.

[0016] Furthermore, construct a 3-degree-of-freedom nonlinear mathematical model of the rotary sailboat as shown in formulas (1) and (2),

[0017]

[0018]

[0019] where η = [x, y, ψ] TDenote the position coordinates and heading angle of the junk rig sailboat in the earth-fixed coordinate system; ν = [u, v, r] T Denote the forward, lateral drift and yaw velocities of the junk rig sailboat in the appended body coordinate system; m u ,m v ,m r Denote the added mass of the junk rig sailboat in the forward, lateral drift and yaw degrees of freedom; d wu ,d wr ,d wv Denote the unknown ocean environmental disturbances of the junk rig sailboat in the forward, lateral drift and yaw degrees of freedom; f u (v), f v (v), f r (v) denotes the model uncertainty terms of the junk rig sailboat in the forward, lateral drift and yaw degrees of freedom; F p = T u (·)|n|n and M r = F r (·)δ respectively denote the forward thrust and yaw moment generated by the propeller and rudder, where T u (·) denotes the main engine gain function, F r (·) denotes the servo gain function, n denotes the main engine speed, and δ denotes the rudder angle; d u1 ,d u2 ,d u3 ,d v1 ,d v2 ,d v3 ,d r1 ,d r2 ,d r3 Denote the unknown hydrodynamic parameters; F s Denote the auxiliary thrust generated by the junk sail, as shown in formula (3),

[0020]

[0021] where, ρ A Denote the air density, V a Denote the apparent wind speed, A denotes the maximum projected area of the junk sail; C L ,C D Denote the lift and drag coefficients of the junk sail respectively, and β denotes the angle between the apparent wind direction and the ship's heading.

[0022] Furthermore, S2: Set waypoints, use LVS to generate a virtual reference path based on the waypoints, and obtain the guidance law of the junk rig sailboat and the position error of the junk rig sailboat to LVS according to the virtual reference path; Improve the guidance law of the junk rig sailboat using the finite boundary circle rule to obtain the improved ILVS guidance law, and obtain the heading error of the junk rig sailboat to LVS based on the improved ILVS guidance law, including:

[0023] S21. Set waypoints and generate a virtual reference path based on the waypoints, as shown in Equation (4).

[0024]

[0025] In the formula, (x d , y d ) represents the position coordinates of the LVS, ψ d represents the heading angle of the LVS, u d represents the desired speed of the LVS, r d = u d / R i , r d represents the desired turning angular velocity of the LVS, and R i represents the turning radius at the i-th waypoint; the line connecting different position coordinates of the LVS is the virtual reference path.

[0026] S22. Obtain the guidance law of the sloop based on the virtual reference path, as shown in Equation (5).

[0027]

[0028] Among them, (x d , y d ) represents the position coordinates of the LVS, x and y represent the actual ship, that is, the position coordinates of the sloop in the earth-fixed coordinate system, and ψ r represents the guidance law of the sloop.

[0029] S23. Improve the guidance law of the sloop using the finite boundary circle rule to obtain the improved ILVS guidance law, as shown in Equation (6).

[0030]

[0031] In the formula, z e represents the position error between the sloop and the LVS, l min represents the radius of the finite boundary circle; t k represents the time when the guidance law is triggered and updated for the k-th time, and x e , y e represent the distance errors between the sloop and the LVS in the forward and sway degrees of freedom, as shown in Equation (7).

[0032]

[0033] S24. Introduce a dynamic feedback evaluation rule to intervene in the guidance law, and the intervened guidance law is as shown in Equation (8).

[0034]

[0035] where, ψ e = ψ rI - ψ, ψ e represents the heading error between the heading angle of the rotary-sail boat and the desired heading angle, that is, the heading error from the rotary-sail boat to the LVS. I represents the dynamic feedback parameter, and δ sat is the saturation threshold related to the rudder angle, and ψ rI represents the guidance law after intervention.

[0036] Furthermore, virtual control laws for the position error and the heading error are designed based on the actor-critic neural network, including:

[0037] S31. Differentiate the position error and the heading error of the rotary-sail boat, as shown in formula (9),

[0038]

[0039] where, z Δ represents the intermediate variable of the derivative of the position error, as shown in formula (10),

[0040]

[0041] S32. Design virtual control laws for the position error and the heading error based on the actor-critic neural network, as shown in formulas (11) and (12),

[0042]

[0043] where, α u , α r are the virtual control laws for the position error and the heading error respectively, and k z , k ψ are design parameters; W au1 , W ar1 represent the weights of the actor neural network on the forward speed and the heading speed in the virtual control laws for the position error and the heading error respectively, represent the estimated values of W au1 , W ar1 respectively; S u1 , S r1 represent the Gaussian functions in the virtual control system on the displacement speed and the displacement direction; the adaptation rates of the weights of the actor neural network on the forward speed and the heading speed are as shown in formula (13),

[0044]

[0045] where, k au1 , k ar1are design parameters; W cu1 , W cr1 respectively represent the critic neural network weights in the displacement velocity and displacement direction in the virtual control law, respectively represent W cu1 , W cr1 estimates of; the adaptation rates of the critic neural network weights in the displacement velocity and displacement direction As shown in formula (14),

[0046]

[0047] where k cu1 , k cr1 are positive design parameters, u e , r e respectively represent the dynamic errors in the displacement velocity and displacement direction.

[0048] Furthermore, the dynamic surface control technology is introduced to filter the virtual control laws of the position error and the heading error respectively. Based on the filtered virtual control laws, the dynamic errors are defined, and the robust neural damping technology is used to robustify the non - linear terms in the dynamic errors, obtaining the robustified dynamic errors, including:

[0049] The dynamic surface control technology is introduced to filter the virtual control law, and the filtering process is as shown in formula (15),

[0050]

[0051] where σ u , σ r are time constants, β u , β r respectively represent the dynamic surface signals of the position error and the heading error, that is, the filtered virtual control laws; the differences between the filtered virtual control laws of the position error and the heading error and the unfiltered virtual control laws of the position error and the heading error are denoted as d u and d r , d u is the dynamic surface error of the displacement velocity, d u =α u -β u , d r represents the dynamic surface error in the displacement direction, d r =α r -β r ;

[0052] Define the dynamic errors in the displacement velocity and displacement direction, as shown in formula (16),

[0053]

[0054] Wherein, u e , r e respectively represent the dynamic errors in the displacement velocity and displacement direction;

[0055] Derive the dynamic errors in the displacement velocity and displacement direction, as shown in formula (17),

[0056]

[0057] Wherein, F s represents the auxiliary thrust generated by the rotary sail, f u (v) and f r (v) represent the model uncertainties of the rotary sailboat in the forward and yaw degrees of freedom;

[0058] For the nonlinear terms f u (v) and f r (v) in the derivative of the dynamic error, perform robustification using the robust neural damping technique, as shown in formula (18),

[0059]

[0060] Wherein, A u , A r represent the neural network weights, S(v) represents the Gaussian function, ε u (v), ε r (v) represent the approximation errors, b u , b r represent the norm values of A u , A r , w u , w r represent the normalized values of A u , A r , ν represents the set of input vectors, ν = [u, v, r];

[0061] Define the robust neural damping terms n u , n r , as shown in formula (19),

[0062]

[0063] Wherein, and respectively represent the upper bounds of the approximation errors ε u (v), ε r (v), and respectively represent the upper bounds of d wu , d wr ;

[0064] represent unknown parameters, represent the damping terms in displacement velocity and direction respectively; ‖·‖ F represents the Frobenius norm;

[0065] The derivative of the robustified dynamic error is obtained from the robust neural damping term, as shown in Equation (20),

[0066]

[0067] where F p represents the forward thrust generated by the propeller and rudder, and M r represents the yawing moment generated by the propeller and rudder;

[0068] Integrate the derivative of the robustified dynamic error to obtain the robustified dynamic error u e1 and r e1 .

[0069] Furthermore, combine the actor-critic neural network, the robustified dynamic error, and the derivative of the filtered virtual control law to design the intermediate control inputs in displacement velocity and displacement direction, including:

[0070] Let τ u = F p + F s , τ r = M r , combine the actor-critic neural network, the robustified dynamic error, and the derivative of the filtered virtual control law to design the intermediate control inputs τ u and τ r , as shown in Equation (21),

[0071]

[0072] where u e1 and r e1 represent the robustified dynamic error, k u , k r , k un , k rn are controller design parameters; and represent the derivatives of the filtered virtual control law; represents the positive parameter combined using the robust neural damping technique; W au2 , W ar2 represent the actor-critic neural network weights in displacement velocity and displacement direction in the actual control system respectively, respectively represent W au2 , W ar2 's estimated value, S u2 , S r2 represents the Gaussian function in the displacement speed and displacement direction in the actual control system, and the adaptation rate of the actor-critic neural network weights in the displacement speed and displacement direction in the actual control system As shown in formula (22),

[0073]

[0074] In the formula, k au2 , k ar2 are positive design parameters, W cu2 , W cr2 respectively represent the critic neural network weights in the displacement speed and displacement direction in the actual control system, respectively represent W cu2 , W cr2 's estimated value, and the adaptation rate of the critic neural network weights in the displacement speed and displacement direction in the actual control system As shown in formula (23),

[0075]

[0076] In the formula, k cu2 , k cr2 are positive design parameters.

[0077] Furthermore, an integral dynamic event-triggering mechanism is designed, that is, formula (24),

[0078]

[0079] In the formula, τ u (t u ), τ r (t r ) respectively represent the intermediate control variables in the displacement speed and displacement direction, τ ku (t ku ), τ kr (t kr ) represents the k-th trigger control command of τ u (t u ), τ r (t r ), and the change error of the intermediate control input in the displacement speed is e u =τ ku -τ u , and the change error of the intermediate control input in the direction is e r =τ kr -τ r , ρu , ρ r represents the self - update threshold parameter, sigmod represents the activation function, c u , c r is a positive time constant; t u and t r represents τ u and τ r are the time variables of τ ku and τ kr respectively represent the updated intermediate control inputs in the displacement speed and displacement direction; t ku and t kr represents τ ku and τ kr are the time variables of τ

[0080] Furthermore, combining the adaptive compensation technology, a ship path - tracking controller is designed based on the updated intermediate control input, including:

[0081] Combining the adaptive compensation technology, a ship path - tracking controller is designed based on the updated intermediate control input. As shown in formula (25), the gain adaptation laws in the displacement speed and displacement direction designed are as shown in formula (26).

[0082]

[0083] In the formula, respectively represent the gain design variables, the estimated values of, T u (·) represents the main engine gain function, F r (·) represents the steering gear gain function, represents the initial value of, respectively represent the gain adaptation laws in the displacement speed and displacement direction, Γ u , Γ r , γ u , γ r are positive adaptive law design parameters.

[0084] Beneficial effects: The present invention provides an event - triggered optimization control method for a rotary - sailboat facing the cruise mission of an ocean ranch, having the following advantages:

[0085] Aiming at the problems that the real - time transmission of the guidance reference signal may lead to input saturation and communication redundancy, an improved ILVS guidance law based on a finite - boundary circle is constructed. While taking into account the input saturation problem, the communication load of the guidance system is effectively reduced.

[0086] (2) Aiming at the problems of channel occupancy and actuator wear existing in the actual cruising process of the rotary sailboat, an integral dynamic event-triggering mechanism coupling output error is proposed. Compared with the existing technologies, the proposed event-triggering mechanism does not require manual setting of threshold parameters, can effectively avoid the phenomenon of long-time non-triggering caused by small-amplitude state fluctuations, and ensure the cruising accuracy of the marine ranch.

[0087] (3) Aiming at the problem of the complex design of the existing reinforcement learning algorithms, an optimized backstepping method and a robust neural damping technology are integrated. A virtual control law is designed based on the Actor-critic Neural Networks (AC-NNs), and the nonlinearity in the control law is robustified through the robust damping technology. This process can gradually explore the optimal solutions for the virtual and actual control systems, and has low design complexity and strong robustness, effectively improving the practicality of the controller in the dynamic marine environment. Brief Description of the Drawings

[0088] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0089] Figure 1 It is the method flow chart of an event-triggering optimization control method for a rotary sailboat facing the cruising task of a marine ranch provided by the present invention;

[0090] Figure 2 It is the structural schematic diagram of the rotary sail;

[0091] Figure 3 It is the force analysis diagram of the rotary sailboat;

[0092] Figure 4 It is the improved ILVS guidance principle framework diagram provided by the present invention;

[0093] Figure 5 It is the ship trajectory diagram under the inspection task of the marine ranch;

[0094] Figure 6 It is the path tracking error example diagram of an embodiment of the present invention;

[0095] Figure 7 It is the control command and actual input broken line diagram of an embodiment of the present invention;

[0096] Figure 8 It is the rotary sail thrust and energy optimization rate curve diagram of an embodiment of the present invention;

[0097] Figure 9 This is the path tracking trajectory diagram for comparing the present invention with the prior art;

[0098] Figure 10 This is the schematic diagram of the position and attitude errors for comparing the present invention with the prior art;

[0099] Figure 11 Example diagram of control commands under control methods of different prior arts;

[0100] Figure 12 This is the schematic diagram of the trigger interval when the present invention is compared with the prior art. Detailed implementation manners

[0101] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0102] This embodiment provides an optimized control method for event triggering of a rotary sailboat for ocean ranch cruising tasks, as Figure 1 shown, including:

[0103] S1: Construct a 3-degree-of-freedom nonlinear mathematical model of the rotary sailboat as the subsequent control object;

[0104] S2: Set waypoints, use LVS to generate a virtual reference path based on the waypoints, obtain the guidance law of the rotary sailboat and the position error of the rotary sailboat relative to LVS according to the virtual reference path; improve the guidance law of the rotary sailboat using the finite boundary circle rule to obtain an improved ILVS guidance law, and obtain the heading error of the rotary sailboat relative to LVS based on the improved ILVS guidance law;

[0105] S3: Design virtual control laws for position error and heading error respectively based on the actor-critic neural network;

[0106] S4: Introduce the dynamic surface control technology to filter the virtual control laws for position error and heading error respectively, define the dynamic error based on the filtered virtual control laws, and use the robust neural damping technology to robustify the nonlinear terms in the dynamic error to obtain the robustified dynamic error;

[0107] S5: Combine the actor-critic neural network, the robustified dynamic error, and the derivative of the filtered virtual control law to design intermediate control inputs in the displacement velocity and displacement direction;

[0108] S6: Design an integral dynamic event-triggering mechanism to determine whether the change error of the intermediate control input satisfies the integral dynamic event-triggering mechanism. If it is satisfied, obtain the intermediate control input for updating the next trigger time. If it is not satisfied, maintain the intermediate control input updated at the previous trigger time until a new intermediate control input is obtained at the next trigger time, and then determine whether the new intermediate control input satisfies the integral dynamic event-triggering mechanism;

[0109] S7: Combine the adaptive compensation technology and design a ship path tracking controller based on the updated intermediate control input. The ship path tracking controller is used to perform path tracking control on the sloop to achieve the autonomous cruising task of the sloop in the marine ranch.

[0110] Specifically, first, construct a 3-degree-of-freedom nonlinear mathematical model of the sloop as the subsequent control object. By establishing the 3-degree-of-freedom nonlinear mathematical model, the dynamic behavior of the sloop can be accurately described, providing an accurate dynamic basis for the subsequent controller design;

[0111] Secondly, S2: Set waypoints, use LVS to generate a virtual reference path based on the waypoints, and obtain the guidance law of the sloop and the position error of the sloop relative to LVS according to the virtual reference path; improve the guidance law of the sloop using the finite boundary circle rule to obtain an improved ILVS guidance law, and obtain the heading error of the sloop relative to LVS based on the improved ILVS guidance law. The virtual reference path generated by LVS can provide a smooth and trackable sailing trajectory for the sloop. Using the finite boundary circle to improve the guidance law can effectively reduce the communication load of the guidance system while taking into account the input saturation problem; design virtual control laws for the position error and the heading error respectively based on the actor-critic neural network, introduce the dynamic surface control technology to filter the virtual control laws for the position error and the heading error respectively, define the dynamic error based on the filtered virtual control laws, and use the robust neural damping technology to robustify the nonlinear terms in the dynamic error to obtain the robustified dynamic error. The actor-critic neural network can adaptively adjust the virtual control laws through online learning and optimizing strategies to improve the control accuracy of the position error and the heading error. At the same time, combined with the reinforcement learning mechanism, the system can adapt to the dynamically changing environment and uncertainties, enhancing the robustness of the controller;

[0112] Again, a middle control input in the displacement velocity and displacement direction is designed by combining the actor-critic neural network, the robustified dynamic error, and the derivative of the filtered virtual control law; an integral dynamic event-triggering mechanism is designed to determine whether the change error of the middle control input satisfies the integral dynamic event-triggering mechanism. If it is satisfied, the middle control input at the next triggering time is obtained. If it is not satisfied, the middle control input at the previous triggering time is maintained until a new middle control input is obtained at the next triggering time, and then it is determined whether the new middle control input satisfies the integral dynamic event-triggering mechanism. Compared with the prior art, the event-triggering mechanism does not require manual setting of threshold parameters, can effectively avoid the phenomenon of long-term non-triggering caused by small-amplitude state fluctuations, ensure the cruising accuracy of the ocean ranch, and perform robustification processing on the non-linearity in the control law through the robust damping technology. This process can gradually explore the optimal solution for the virtual and actual control systems, effectively improving the practicability of the controller in the dynamic ocean environment;

[0113] Finally, in combination with the adaptive compensation technology, a ship path tracking controller is designed based on the updated middle control input. The finally designed ship path tracking controller can achieve high-precision autonomous cruising of the ketch in a complex ocean environment, meeting the actual application requirements of the ocean ranch.

[0114] In a specific embodiment, a 3-degree-of-freedom nonlinear mathematical model of the ketch is constructed, as shown in Formulas (27) and (28),

[0115]

[0116] where η = [x, y, ψ] T represents the position coordinates and heading angle of the ketch in the earth-fixed coordinate system; v = [u, v, r] T represents the forward, sway, and yaw velocities of the ketch in the body-fixed coordinate system; m u , m v , m r represent the added masses of the ketch in the forward, sway, and yaw degrees of freedom; d wu , d wn , d wr represent the unknown ocean environmental disturbances of the ketch in the forward, sway, and yaw degrees of freedom; f u (v), f v (v), f r (v) represent the model uncertainties of the ketch in the forward, sway, and yaw degrees of freedom; F p = T u (·)|·|n and M r = F r (·)δ respectively represent the forward thrust and yaw moment generated by the propeller and rudder, where T u(·) represents the main engine gain function, F r (·) represents the servo gain function, n represents the main engine speed, and δ represents the rudder angle; d u1 ,d u2 ,d u3 ,d v1 ,d v2 ,d v3 ,d r1 ,d r2 ,d r3 represents unknown hydrodynamic parameters; F s represents the auxiliary thrust generated by the rotating cylinder sail, as shown in formula (29),

[0117]

[0118] where, ρ A represents the air density, V a represents the apparent wind speed, A represents the maximum projected area of the rotating cylinder sail; C L ,C D respectively represent the lift and drag coefficients of the rotating cylinder sail, and β represents the angle between the apparent wind direction and the ship's head direction; the structure of the rotating cylinder sail is as Figure 2 shown. In the figure, h represents the height of the rotating cylinder sail, V wind represents the true wind speed, d e represents the diameter of the upper top plate of the rotating cylinder sail, and d represents the diameter of the lower plate; Figure 3 is the force analysis diagram of the rotating cylinder sailboat. In the figure, V t is the same as V wind , both representing the true wind speed, V s is the ship speed; γ is the angle between the true wind speed and the ship's head direction, x b ,y b is the attached body coordinate system of the ship; F l is the side thrust generated by the rotating cylinder sail, and L and D respectively represent the lift and drag generated by the rotating cylinder sail.

[0119] Constructing a 3-degree-of-freedom nonlinear mathematical model of the rotating cylinder sailboat can more accurately describe the dynamic characteristics of the rotating cylinder sailboat, improve the control accuracy and robustness, and provide data support for subsequent virtual control laws and controllers.

[0120] In a specific embodiment, the scheme of setting waypoints, using LVS to generate a virtual reference path based on the waypoints, obtaining the guidance law of the rotating cylinder sailboat and the position error of the rotating cylinder sailboat to LVS according to the virtual reference path; improving the guidance law of the rotating cylinder sailboat using the finite boundary circle rule to obtain an improved ILVS guidance law, and obtaining the heading error of the rotating cylinder sailboat to LVS based on the improved ILVS guidance law is as follows:

[0121] S21, as Figure 4 shown,Figure 4 shows the framework of the improved ILVS guidance principle. In navigation practice, the reference path is usually generated by a series of waypoints set by humans, such as W1, W2, W3..., W n , therefore, the present invention uses a Logical Virtual Ship (LVS) to generate a virtual reference path based on waypoints, as shown in formula (30),

[0122]

[0123] In the formula, (x d , y d ) represents the position coordinates of the LVS, ψ d represents the heading angle of the LVS, u d represents the desired speed of the LVS, r d = u d / R i , r d represents the desired turning angular velocity of the LVS, R i represents the turning radius at the i-th waypoint, and the value of R i shall not be less than the minimum turning radius of a sloop; the line formed by different position coordinates of the LVS is the virtual reference path;

[0124] S22. The guidance law of the sloop is obtained based on the virtual reference path as shown in formula (31),

[0125]

[0126] wherein, (x d , y d ) represents the position coordinates of the LVS, x, y represent the actual ship, that is, the position coordinates of the sloop in the Earth-fixed coordinate system, and ψ r represents the guidance law of the sloop;

[0127] S23. The finite boundary circle rule is used to improve the guidance law of the sloop to obtain the improved ILVS guidance law, as shown in formula (32),

[0128]

[0129] The finite boundary circle is as Figure 4 shown. In the formula, z e represents the position error of the sloop relative to the LVS, l min represents the radius of the finite boundary circle; t k represents the time when the guidance law is triggered and updated for the k-th time, x e , y eIndicates the distance error between the rotary sailboat and the LVS in the forward and cross-drift degrees of freedom, as shown in Equation (33).

[0130]

[0131] S24. Considering the problem of input saturation in the control system, a dynamic feedback evaluation rule is introduced to further intervene in the guidance law. The intervened guidance law is shown in Equation (34).

[0132]

[0133] where, ψ e = ψ rI - ψ, ψ e represents the heading error between the heading angle of the rotary sailboat and the desired heading angle, that is, the heading error from the rotary sailboat to the LVS. I represents the dynamic feedback parameter, and δ sat is the saturation threshold related to the rudder angle. ψ rI represents the intervened guidance law.

[0134] In this scheme, the finite boundary circle is used to improve the guidance law, and an improved ILVS guidance law based on the finite boundary circle is constructed. While taking into account the problem of input saturation, it effectively reduces the communication load of the guidance system and solves the problems of input saturation and communication redundancy that may be caused by the real-time transmission of the guidance reference signal.

[0135] In a specific embodiment, the scheme for separately designing the virtual control laws for the position error and the heading error based on the actor-critic neural network is as follows:

[0136] S31. Differentiate the position error and the heading error of the rotary sailboat, as shown in Equation (35).

[0137]

[0138] In the formula, z Δ represents the intermediate variable of the derivative of the position error, as shown in Equation (36).

[0139]

[0140] S32. To stabilize the position error and the heading error, virtual control laws for the position error and the heading error are separately designed based on the actor-critic neural network, as shown in Equations (37) and (38).

[0141]

[0142] In the formula, α u , α r are respectively the virtual control laws for the position error and the heading error, and kz , k ψ is a positive design parameter; W au1 , W ar1 In the virtual control laws representing the position error and the heading error respectively, the weights of the actor neural network on the forward speed and the heading speed respectively represent W au1 , W ar1 's estimated value; S u1 , S r1 represents the Gaussian functions in the displacement speed and the displacement direction in the virtual control system; the adaptation rate of the weights of the actor neural network on the forward speed and the heading speed As shown in formula (39),

[0143]

[0144] where k au1 , k ar1 is a positive design parameter; W cu1 , W cr1 respectively represent the weights of the critic neural network on the displacement speed and the displacement direction in the virtual control law, respectively represent W cu1 , W cr1 's estimated value; the adaptation rate of the weights of the critic neural network on the displacement speed and the displacement direction As shown in formula (40),

[0145]

[0146] where k cu1 , k cr1 is a positive design parameter, u e , r e respectively represent the dynamic errors on the displacement speed and the displacement direction.

[0147] In this solution, the virtual control laws for the position error and the heading error are respectively designed based on the actor-critic neural network, which can gradually explore the optimal solutions for the virtual and actual control systems, and has a low design complexity, effectively improving the practicability of the controller in the dynamic ocean environment and solving the problem of the complex design of the existing reinforcement learning algorithms.

[0148] In a specific embodiment, the dynamic surface control technology is introduced to filter the virtual control laws for the position error and the heading error respectively, the dynamic errors are defined based on the filtered virtual control laws, and the robust neural damping technology is used to robustify the non-linear terms in the dynamic errors. The solution to obtain the robustified dynamic errors is as follows:

[0149] The dynamic surface control technology (DSC) is introduced to filter the virtual control law to avoid repeated differentiation of the virtual control laws α u and α r so as to prevent the occurrence of the "computational explosion" phenomenon and simplify the subsequent controller design. The filtering process is shown in Equation (41).

[0150]

[0151] where σ u , σ r are time constants greater than zero, β u , β r represent the dynamic surface signals of the position error and the heading error respectively, that is, the filtered virtual control laws; the differences between the virtual control laws of the filtered position error and heading error and the virtual control laws of the position error and heading error before filtering are denoted as d u and d r , d u is the dynamic surface error of the displacement velocity, d u =α u -β u , d r represents the dynamic surface error in the displacement direction, d r =α r -β r ;

[0152] Define the dynamic errors in the displacement velocity and displacement direction as shown in Equation (42).

[0153]

[0154] where u e , r e represent the dynamic errors in the displacement velocity and displacement direction respectively;

[0155] Differentiate the dynamic errors in the displacement velocity and displacement direction as shown in Equation (43).

[0156]

[0157] where F s represents the auxiliary thrust generated by the rotating cylinder sail, f u (v) and f r (v) represent the model uncertainties of the rotating cylinder sailboat in the surge and yaw degrees of freedom;

[0158] For the nonlinear terms f u (v) and f r(v) is robustified using the robust neural damping technique, as shown in Equation (44).

[0159]

[0160] where A u , A r represents the neural network weights, S(v) represents the Gaussian function, ε u (v), ε r (v) represents the approximation error, b u , b r represents the norm value of A u , A r , w u , w r represents the normalized value of A u , A r , ν represents the set of input vectors, ν = [u, v, r];

[0161] Define the robust neural damping term n u , n r , as shown in Equation (45).

[0162]

[0163] where and respectively represent the upper bounds of the approximation errors ε u (v), ε r (ν), and respectively represent the upper bounds of d wu , d wr ;

[0164] represents the unknown parameter, respectively represent the damping terms in the displacement velocity and direction; ‖·‖ F represents the Frobenius norm;

[0165] According to the robust neural damping term, the derivative of the robustified dynamic error is obtained, as shown in Equation (46).

[0166]

[0167] where F p represents the forward thrust generated by the propeller and rudder, M r represents the yawing moment generated by the propeller and rudder;

[0168] Integrate the derivative of the robustified dynamic error to obtain the robustified dynamic errors u e1 and r e1。

[0169] In this solution, the dynamic surface control technology (DSC) is introduced to filter the virtual control law to avoid repeated differentiation of the virtual control laws α u and α r , thus preventing the occurrence of the "computational explosion" phenomenon and simplifying the subsequent controller design. The robust neural damping technology is used to robustify the non-linear terms in the dynamic error, which can enhance the robustness, stability and control accuracy of the system, and effectively improve the practicality of the controller in the dynamic ocean environment.

[0170] In a specific embodiment, the scheme for designing the intermediate control inputs in the displacement velocity and displacement direction by combining the actor-critic neural network, the robustified dynamic error and the derivative of the filtered virtual control law is as follows:

[0171] Let τ u = F p + F s , τ r = M r . Combine the actor-critic neural network, the robustified dynamic error and the derivative of the filtered virtual control law to design the intermediate control inputs τ u and τ r , as shown in formula (47).

[0172]

[0173] In the formula, u e1 and r e1 represent the robustified dynamic error, k u , k r , k un , k rn are positive controller design parameters; and represent the derivative of the filtered virtual control law; represents the positive parameters combined using the robust neural damping technology; W au2 , W ar2 respectively represent the actor-critic neural network weights in the displacement velocity and displacement direction in the actual control system. respectively represent the estimated values of W au2 , W ar2 . S u2 , S r2 represent the Gaussian functions in the displacement velocity and displacement direction in the actual control system. The adaptation rates of the actor-critic neural network weights in the displacement velocity and displacement direction in the actual control system As shown in formula (48),

[0174]

[0175] where k au2 , k ar2 are positive design parameters, and W cu2 , W cr2 respectively represent the critic neural network weights in the displacement velocity and displacement direction in the actual control system, respectively represent the estimated values of W cu2 , W cr2 , and the adaptation rates of the critic neural network weights in the displacement velocity and displacement direction in the actual control system As shown in formula (49),

[0176]

[0177] where k cu2 , k cr2 are positive design parameters.

[0178] In this solution, by combining the actor-critic neural network, the robustified dynamic error, and the derivative of the filtered virtual control law to design the intermediate control input, it can improve the control accuracy, enhance the robustness, and improve the dynamic response. At the same time, designing the intermediate control input can provide data support for the design of the controller.

[0179] In a specific embodiment, an integral dynamic event-triggering mechanism is designed to determine whether the change error of the intermediate control input satisfies the integral dynamic event-triggering mechanism. If it is satisfied, the intermediate control input at the next triggering time is obtained. If it is not satisfied, the intermediate control input at the previous triggering time is maintained until a new intermediate control input is obtained at the next triggering time, and then it is determined whether the new intermediate control input satisfies the integral dynamic event-triggering mechanism. The solution is as follows:

[0180] An integral dynamic event-triggering mechanism is proposed to reduce the frequent command updates between the controller and the actuator and ensure good triggering performance when the system state fluctuates slightly, that is, formula (50),

[0181]

[0182] where τ u (t u ), τ r (t r ) respectively represent the intermediate control variables in the displacement velocity and displacement direction, and τ ku (t ku ), τ kr (t kr ) represents τu (t u ), τ r (t r ) The k-th trigger control command, and the intermediate control input change error in displacement speed is e u = τ ku - τ u , and the intermediate control input change error in direction is e r = τ kr - τ r , ρ u , ρ r represents the self-update threshold parameter, sigmod represents the activation function, c u , c r is a positive time constant; t u and t r represent the time variables of τ u and τ r ; τ ku and τ kr respectively represent the updated intermediate control inputs in displacement speed and displacement direction; t ku and t kr represent the time variables of τ ku and τ kr .

[0183] In this solution, an integral dynamic event-triggering mechanism for coupled output error is proposed. Compared with the prior art, the proposed event-triggering mechanism does not require manual setting of threshold parameters, can effectively avoid the phenomenon of long-term non-triggering caused by small-amplitude state fluctuations, and ensure the cruising accuracy of the ocean ranch.

[0184] In a specific embodiment, combining the adaptive compensation technology, a ship path tracking controller is designed based on the updated intermediate control input. The solution for the ship path tracking controller to perform path tracking control on the ketch to achieve the autonomous cruising task of the ketch in the ocean ranch is:

[0185] Combining the adaptive compensation technology, a ship path tracking controller is designed based on the updated intermediate control input. As shown in formula (51), the gain adaptive laws in displacement speed and displacement direction are designed as shown in formula (52),

[0186]

[0187] Wherein, respectively represent the gain design variables, the estimated value of, T u (·) represents the main engine gain function, F r (·) represents the servo gain function, represents The initial value of respectively represents the gain adaptation laws in the displacement speed and displacement direction. Γ u , Γ r , γ u , γ r are positive adaptive law design parameters;

[0188] The ship path tracking controller is used to perform path tracking control on the junk rig sailboat, so as to achieve the autonomous cruising task of the junk rig sailboat in the marine ranch.

[0189] In this solution, the designed ship path tracking controller can effectively improve the path tracking control accuracy of the junk rig sailboat and ensure the cruising accuracy of the marine ranch.

[0190] To verify the effectiveness and superiority of the present invention, the underactuated ship model in Fossen's underactuated ship model and the junk rig sail model in Tillig's junk rig sail model are selected as the control objects. At the same time, based on the NORSOK wind spectrum and the JONSWAP wave spectrum, a mathematical model based on physical principles is constructed to simulate the real marine environment during the inspection of the marine ranch. A smooth reference path is generated through the improved ILVS guidance principle, and the desired speed u d = 3.5 m / s.

[0191] Figure 5 shows the ship's track under the inspection task of the marine ranch. By adopting the improved ILVS guidance principle and the reinforcement learning control method, the ship can adapt to complex inspection routes and effectively avoid overshoot;

[0192] The path tracking error results obtained using the present invention are as Figure 6 shown. The algorithm mentioned in the figure is the method provided by the present invention;

[0193] Figure 7 represents the control command and the actual input obtained by the present invention. By integrating the main engine and the servo system of the rudder, mechanical damage caused by actuator overload can be effectively prevented, thereby extending its service life.

[0194] Figure 8 shows the thrust of the junk rig sail (Energy Optimization Rate, EOR) and its curve, where EOR(t) = F s (t) / (F p (t) + F s (t)). The calculation results show that the use of the junk rig sail auxiliary technology reduces the energy consumption of the entire voyage by approximately 14.1%.

[0195] Figures 9 - 12Shows the comparison experiment results of the proposed event-triggered algorithm - the method with an integral event-triggering mechanism in the present invention, the proposed continuous algorithm - the method without an integral event-triggering mechanism in the present invention, and the event-triggered approximate optimal path tracking control algorithm with state constraints:

[0196] Figure 9 Shows the tracking trajectories of the two methods and the existing technology. From Figure 9 it can be seen that the tracking trajectories of the three methods can all converge well to the reference path, while ensuring that the lateral error remains within ±1m;

[0197] Figure 10 Shows the ship position and attitude errors using the two methods and the existing technology. The results show that the two methods proposed in the present invention are significantly superior to the event-triggered approximate optimal path tracking control algorithm with state constraints in terms of performance. It is worth noting that the control accuracy of the proposed event-triggered algorithm is extremely close to that of the continuous algorithm, fully demonstrating the rationality and superiority of the proposed integral dynamic event-triggering mechanism.

[0198] Figure 11 and Figure 12 respectively show the control command curves and trigger interval curves under different methods. Due to the superiority of the event-triggering mechanism proposed in the present invention, its control command curve is smoother and the average trigger interval is longer, effectively reducing the chattering of the control signal and avoiding excessive wear of the actuator.

[0199] Combined with the existing technology, guidance law construction, controller design, and simulation experiment comparison, the present invention has the following three significant beneficial effects in the path tracking of sail-type ships:

[0200] 1) Combining the existing virtual guidance algorithm, an improved ILVS guidance principle is proposed. While taking into account the input saturation problem, the introduction of the finite boundary circle rule effectively reduces the transmission frequency of the guidance signal and avoids the occurrence of communication redundancy in the guidance system.

[0201] 2) Considering the ocean engineering practice, the present invention designs an integral dynamic event-triggering mechanism that couples the output error. This event-triggering mechanism does not require manual setting of threshold parameters and can avoid the decline in control accuracy caused by long-term non-triggering, effectively solving the problems of channel occupancy and actuator wear in the actual cruising process of ocean pastures.

[0202] 3) Based on the actor-critic neural network, the present invention proposes a robust optimization control method of reinforcement learning with low design complexity, effectively avoiding the decline in system response ability caused by controller processing delay, which is of great significance for ensuring the safe navigation of the rotary sailboat at sea.

[0203] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An event-triggered optimization control method for a rotary sailboat for marine ranch cruising missions, characterized in that: include: S1: Construct a 3-DOF nonlinear mathematical model of the rotary sailboat as the subsequent control object; S2: Set waypoints, use LVS to generate a virtual reference path based on the waypoints, obtain the guidance law of the drum sailboat and the position error from the drum sailboat to the LVS according to the virtual reference path; use the finite boundary circle rule to improve the guidance law of the drum sailboat to obtain the improved ILVS guidance law, and obtain the heading error from the drum sailboat to the LVS based on the improved ILVS guidance law; S3: Design virtual control laws for position error and heading error based on actor-critic neural network; S4: Dynamic surface control technology is introduced to filter the virtual control laws of position error and heading error respectively, and the dynamic error is defined based on the filtered virtual control law. The robust neural damping technology is used to robustly process the nonlinear terms in the dynamic error to obtain the robustly processed dynamic error. S5: Design of intermediate control inputs in displacement velocity and displacement direction by combining the actor-critic neural network, robustified dynamics error and the derivative of the filtered virtual control law; S6: Design an integral dynamic event trigger mechanism to determine whether the change error of the intermediate control input satisfies the integral dynamic event trigger mechanism. If so, obtain the intermediate control input of the next trigger time. If not, keep the intermediate control input of the previous trigger time until a new intermediate control input is obtained at the next trigger time, and then determine whether the new intermediate control input satisfies the integral dynamic event trigger mechanism. S7: In combination with the adaptive compensation technology, a ship path tracking controller is designed based on the updated intermediate control input. The ship path tracking controller is used to perform path tracking control on the rotary sailboat to achieve the autonomous cruising mission of the rotary sailboat in the marine ranch.

2. The event-triggered optimization control method for a rotary sailboat for marine ranching cruising mission according to claim 1 is characterized in that: The three-degree-of-freedom nonlinear mathematical model of the rotary sailboat is constructed as shown in formulas (1) and (2). Where η = [x, y, ψ] T represents the position coordinates and heading angle of the rotary sailboat in the earth-fixed coordinate system; v = [u, v, r] T represents the forward, drifting and bow rolling speeds of the rotary sailboat in the appendage coordinate system; m u ,m v ,m r It represents the additional mass of the rotary sailboat in the forward, sideways and bow pitching degrees of freedom; d wu ,d wr ,d wv represents the unknown ocean environment disturbance of the rotary sailing vessel in terms of forward, sideways and bow pitching degrees of freedom; f u (v),f v (v),f r (v) represents the model uncertainty of the rotary sailboat in terms of forward, sideways and pitching degrees of freedom; F p =T u (·)|·|n and M r =F r (·)δ represents the forward thrust and bow moment generated by the propeller and rudder, respectively, where T u (·) represents the host gain function, F r (·) represents the steering gear gain function, n represents the main engine speed, δ represents the rudder angle; d u1 ,d u2 ,d u3 ,d v1 ,d v2 ,d v3 ,d r1 ,d r2 ,d r3 represents the unknown hydrodynamic parameters; F s represents the auxiliary thrust generated by the rotary sail, as shown in formula (3), Among them, ρ A Represents the air density, V a represents the apparent wind speed, A represents the maximum projected area of ​​the rotary sail; C L ,C D They represent the lift and drag coefficients of the drum sail respectively, and β represents the angle between the apparent wind direction and the bow direction of the ship.

3. The event-triggered optimization control method for a rotary sailboat for marine ranching cruising mission according to claim 2 is characterized in that: Set waypoints, use LVS to generate a virtual reference path based on the waypoints, obtain the guidance law of the drum sailboat and the position error from the drum sailboat to the LVS according to the virtual reference path; use the finite boundary circle rule to improve the guidance law of the drum sailboat to obtain the improved ILVS guidance law, and obtain the heading error from the drum sailboat to the LVS based on the improved ILVS guidance law; including: S21, setting waypoints, and generating a virtual reference path based on the waypoints, as shown in formula (4), In the formula, (x d ,y d ) represents the position coordinates of LVS, ψ d represents the heading angle of LVS, u d represents the expected speed of LVS, r d =u d / R i , represents the desired bow angular velocity of LVS, R i represents the turning radius at the i-th waypoint; the line connecting the different position coordinates of LVS is the virtual reference path; S22. The guidance law of the rotary sailboat is obtained based on the virtual reference path as shown in formula (5): Among them, (x d ,y d ) represents the position coordinates of LVS, x, y represents the position coordinates of the actual ship, i.e. the rotary sailboat in the earth fixed coordinate system, ψ r represents the guidance law of a rotary sailing vessel; S23. The guidance law of the rotary sailboat is improved by using the finite boundary circle rule to obtain the improved ILVS guidance law, as shown in formula (6): In the formula, z e It represents the position error from the drum sailboat to the LVS, l min represents the radius of the finite boundary circle; t k represents the time when the guidance law triggers the update for the kth time, x e ,y e It represents the distance error between the drum sailboat and the LVS in the forward and horizontal drift degrees of freedom, as shown in formula (7). S24, introduce dynamic feedback evaluation rules to intervene in the guidance law. The guidance law after intervention is shown in formula (8): Among them, ψ e =ψ rI -ψ, ψ e represents the heading error between the heading angle of the rotary sailboat and the desired heading angle, that is, the heading error from the rotary sailboat to the LVS, I represents the dynamic feedback parameter, δ sat is the saturation threshold related to the rudder angle, ψ rI represents the guidance law after intervention.

4. The event-triggered optimization control method for a rotary sailboat for marine ranching cruising mission according to claim 3 is characterized in that: The virtual control laws for position error and heading error are designed based on the actor-critic neural network, including: S31. Deriving the position error and heading error of the rotary sailboat, as shown in formula (9), In the formula, z Δ represents the intermediate variable for the position error derivative, as shown in formula (10), S32. Based on the actor-critic neural network, the virtual control laws for position error and heading error are designed respectively, as shown in formulas (11) and (12), In the formula, α u ,α r are the virtual control laws for position error and heading error, respectively, k z ,k ψ is the design parameter; W au1 ,W ar1 The weights of the actor neural network on the forward speed and heading speed in the virtual control law representing the position error and heading error, respectively, Respectively represent W au1 ,W ar1 The estimated value of S u1 ,S r1 represents the Gaussian function of displacement speed and displacement direction in the virtual control system; the adaptation rate of the actor neural network weights on the forward speed and heading speed As shown in formula (13), In the formula, k au1 ,k ar1 is the design parameter; W cu1 ,W cr1 denote the critic neural network weights of displacement velocity and displacement direction in the virtual control law, respectively. Respectively represent W cu1 ,W cr1 Estimated value of displacement speed and displacement direction; Adaptation rate of critic neural network weights As shown in formula (14), In the formula, k cu1 ,k cr1 is a positive design parameter, u e ,r e represent the dynamic errors in displacement velocity and displacement direction respectively.

5. The event-triggered optimization control method for a rotary sailboat for marine ranching cruising missions according to claim 4 is characterized in that: The dynamic surface control technology is introduced to filter the virtual control laws of the position error and heading error respectively. The dynamic error is defined based on the filtered virtual control law, and the robust neural damping technology is used to robustly process the nonlinear terms in the dynamic error. The robustly processed dynamic error is obtained, including: The dynamic surface control technology is introduced to filter the virtual control law. The filtering process is shown in formula (15): In the formula, σ u ,σ r is the time constant, β u ,β r The dynamic surface signals of the position error and heading error are represented by the filtered virtual control law. The difference between the virtual control law of the position error and heading error after filtering and the virtual control law of the position error and heading error before filtering is represented by d u and d r , d u is the dynamic surface error of displacement velocity, d u =α u -β u , d r Represents the dynamic surface error in the displacement direction, d r =α r -β r ; Define the dynamic error in displacement velocity and displacement direction as shown in formula (16): In the formula, u e ,r e represent the dynamic errors in displacement velocity and displacement direction respectively; The dynamic error in displacement velocity and displacement direction is derived as shown in formula (17): In the formula, F s represents the auxiliary thrust generated by the rotor sail, f u (ν) and f r (ν) represents the model uncertainty of the rotary sailboat in the forward and forward pitching degrees of freedom; The nonlinear term f in the dynamic error derivative u (ν) and f r (ν) is robustified using robust neural damping technology, as shown in formula (18), In the formula, A u ,A r represents the neural network weight, S(ν) represents the Gaussian function, ε u (v),ε r (v) represents the approximation error, b u ,b r Indicates A u ,A r The norm value, w u ,w r Indicates A u ,A r The normalized value of , ν represents the set of input vectors, v = [u, v, r]; Define the robust neural damping term n u ,n r , as shown in formula (19), In the formula, and They represent the approximation error ε u (v),ε r The upper bound of (v) is and Respectively represent d wu ,d wr The upper bound of Represents unknown parameters, represent the damping terms of displacement velocity and direction respectively; ‖·‖ F represents the Frobenius norm; According to the robust neural damping term, the derivative of the robust dynamic error is obtained, as shown in formula (20): In the formula, F p is the forward thrust generated by the propeller and rudder, M r represents the bow moment produced by the propeller and rudder; Integrate the derivative of the robust dynamics error to obtain the robust dynamics error u e1 and r e1 .

6. The event-triggered optimization control method for a rotary sailboat for marine ranching cruise missions according to claim 5 is characterized in that: The intermediate control inputs for displacement velocity and displacement direction are designed by combining the actor-critic neural network, robustified dynamics error, and filtered derivatives of the virtual control law, including: Let τ u =F p +F s , τ r =M r , combining the actor-critic neural network, the robustified dynamic error and the derivative of the filtered virtual control law to design the intermediate control input τ in displacement velocity and displacement direction u and τ r , as shown in formula (21), In the formula, u e1 and r e1 represents the robustified dynamic error, k u ,k r ,k un ,k rn is the controller design parameter; and represents the derivative of the filtered virtual control law; represents the positive parameter incorporated using the robust neural damping technique; W au2 ,W ar2 They represent the weights of the actor neural network in displacement speed and displacement direction in the actual control system, Respectively represent W au2 ,W ar2 The estimated value of S u2 ,S r2 represents the Gaussian function of displacement speed and displacement direction in the actual control system, and the adaptive rate of the weight of the actor neural network in the displacement speed and displacement direction in the actual control system As shown in formula (22), In the formula, k au2 ,k ar2 is a positive design parameter, W cu2 ,W cr2 represent the critic neural network weights of displacement velocity and displacement direction in the actual control system, Respectively represent W cu2 ,W cr2 The estimated value of displacement velocity and displacement direction in the actual control system is the adaptive rate of the critic neural network weights. As shown in formula (23), In the formula, k cu2 ,k cr2 is a positive design parameter.

7. The event-triggered optimization control method for a rotary sailboat for marine ranching cruising missions according to claim 6 is characterized in that: Design the integral dynamic event trigger mechanism, that is, formula (24), In the formula, τ u (t u ),τ r (t r ) represent the intermediate control variables of displacement speed and displacement direction, τ ku (t ku ),τ kr (t kr ) represents τ u (t u ),τ r (t r ) The kth trigger control command, the intermediate control input change error on the displacement speed is e u =τ ku -τ u , the intermediate control input change error in the direction is e r =τ kr -τ r , ρ u ,ρ r represents the self-update threshold parameter, sigmod represents the activation function, c u ,c r is a positive time constant; t u and t r Represents τ u and τ r The time variable of ku and τ kr They represent the updated intermediate control inputs of displacement speed and displacement direction respectively; t ku and t kr Represents τ ku and τ kr time variable.

8. The event-triggered optimization control method for a rotary sailboat for marine ranching cruising missions according to claim 7 is characterized in that: Combined with the adaptive compensation technology, the ship path tracking controller is designed based on the updated intermediate control input, including: Combined with the adaptive compensation technology, the ship path tracking controller is designed based on the updated intermediate control input, as shown in formula (25). The designed displacement velocity and displacement direction gain adaptive law is shown in formula (26): In the formula, denote the gain design variables, The estimated value of T u (·) represents the host gain function, F r (·) represents the servo gain function, express The initial value of They represent the gain adaptation law in displacement velocity and displacement direction, Γ u ,Γ r ,γ u ,γ r is a positive adaptive law design parameter.

Images