Under-actuated autonomous water surface vehicle preset time performance control method based on digital twinning

Abstract

The invention discloses a digital twinning-based underactuated autonomous water surface vehicle predetermined time performance control method, which comprises the following steps: establishing an ASV kinetic model of an underactuated autonomous water surface vehicle to construct a digital twinning entity model containing unknown time-varying dynamic items under complex sea conditions, and establishing a digital twinning entity model containing unknown time-varying dynamic items under complex sea conditions on the basis of a long short-term memory (LSTM) network to control the predetermined time performance of the underactuated autonomous water surface vehicle. Obtaining an optimized digital twinning entity model containing an LSTM (Long Short Term Memory) modeling error; constructing an online learning strategy based on a long short-term memory (LSTM) network so as to learn and acquire a time-varying dynamic item in real time; estimating an LSTM modeling error in combination with the constructed meta-learning extended state observer so as to update the digital optimization twin entity model; designing a parallel tracking control law for optimizing synchronous mapping of the digital twin entity model and the ASV kinetic model on the basis of construction of a specified time performance function; according to a parallel tracking control law, an event triggering controller with a relative threshold value is constructed to design an event triggering mechanism based on a memory, and the problems that a traditional ASV control method does not have high control precision and stability under complex sea conditions, consumption of communication resources cannot be effectively reduced, and the service life of the system is prolonged are solved. And the control precision of the under-actuated ASV cannot be ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous surface vehicle (ASV) control, and particularly to a predetermined time performance control method for an underactuated autonomous surface vehicle based on digital twin. Background Art

[0002] Unmanned surface vehicles (USVs) are increasingly widely used in ocean engineering, covering multiple fields such as environmental monitoring, water sampling, and rescue operations. Due to their small size and high maneuverability, USVs show significant advantages when performing complex tasks. Compared with a single USV, a multi-USV formation can perform more complex tasks. Therefore, in the research of USVs, the formation control problem has become an important research direction. However, most of the existing research focuses on the formation control of fully actuated USVs, while the research on underactuated USVs is relatively less. Since the design of the controller for fully actuated USVs is relatively simple, usually three control inputs are required to manage three degrees of freedom, while an underactuated USV only needs two control signals to manage three degrees of freedom, which makes the design of its controller more challenging. In the actual ocean environment, most USVs are underactuated, which means that they need to use two control signals to manage three degrees of freedom. In addition, many studies on underactuated USVs can only achieve the boundedness of the state, which means that the USV may take an infinite time to reach the desired formation shape. In ocean engineering, better performance usually needs to be achieved within a finite time, so the transient performance is often ignored. In fact, having excellent transient performance is usually desirable. Therefore, designing a controller for underactuated USVs is more challenging and important than for fully actuated USVs.

[0003] Currently, the high-precision trajectory tracking of underactuated ASVs needs to break through the following technical bottlenecks:

[0004] 1) Uncertainty of the leader's speed: Most methods do not fully consider the uncertainty of the leader's speed, which mainly stems from the high cost of the sensors required to measure the speed, and ocean communication is limited by the limited bandwidth, and the traditional fixed-interval transmission mechanism is difficult to overcome this limitation; this problem poses an obstacle to the effective cooperative control of USV formations because the speed information of the leader is crucial for the success of the cooperative formation.

[0005] 2) Complex environmental and operating conditions: In the operation of USVs, considering the combined effects of uncertainty, parameter perturbation, underactuation, input saturation, and external interference, the design of USV controllers becomes particularly complex. These factors together pose great challenges to USV formations under different environmental and operating conditions, and require the controller to have high adaptability in order to cope with unknown water characteristics, changing meteorological conditions, continuous changes in system parameters, and limitations of manipulation inputs.

[0006] 3) Performance guarantee within a finite time: Traditional cooperative formation control methods for unmanned boats cannot ensure the smooth operation of unmanned boat formations under specific performance criteria and finite time constraints. Summary of the Invention

[0007] The present invention provides a predetermined-time performance control method for an underactuated autonomous surface vehicle based on digital twin to overcome the above technical problems.

[0008] To achieve the above object, the technical solution of the present invention is as follows:

[0009] A predetermined-time performance control method for an underactuated autonomous surface vehicle based on digital twin specifically includes the following steps:

[0010] S1: Establish an ASV dynamics model of the underactuated autonomous surface vehicle, and construct a digital twin entity model including unknown time-varying dynamic terms under complex sea conditions according to the ASV dynamics model;

[0011] And based on the long short-term memory network LSTM, obtain an optimized digital twin entity model including LSTM modeling errors;

[0012] S2: According to the optimized digital twin entity model, construct an online learning strategy based on the long short-term memory network LSTM to learn in real time and obtain time-varying dynamic terms;

[0013] And combine the constructed meta-learning extended state observer to estimate the LSTM modeling error and time-varying dynamic terms to obtain an ASV dynamics estimation model;

[0014] S3: According to the ASV dynamics estimation model, design a parallel tracking control law based on the constructed specified-time performance function to realize the synchronous mapping of the optimized digital twin entity model and the ASV dynamics model;

[0015] S4: Construct an event-triggered controller with a relative threshold according to the parallel tracking control law;

[0016] And based on the event-triggered controller with a relative threshold, design a memory-based event-triggered mechanism for dynamically adjusting the event-triggering condition to optimize communication resources to achieve the predetermined-time performance control of the underactuated autonomous surface vehicle.

[0017] Further, the S1 specifically includes the following steps:

[0018] S11: Establish an ASV dynamics model of the underactuated autonomous surface vehicle, and its model expression is

[0019]

[0020] where: η represents the position and heading angle of the ASV and η = [x, y, ψ] T ; v represents the velocity vector and v = [u, v, r] T ; u, v, r represent the forward velocity, lateral velocity, and yaw angular velocity; q represents the control input and τ = q = [q u , 0, q r T ; q u , q r represents the control input provided by the thrusters and rudder; M, C(v), D(v) represent the mass matrix, Coriolis matrix, and hydrodynamic damping matrix respectively; d represents the unknown disturbance and d = [d u , d v , d r T ; d u , d v , d r represents the external unknown disturbance of the underactuated ASV; R(ψ) represents the rotation matrix used to transform the hull coordinate system to the inertial coordinate system; represents the first derivative of η; represents the first derivative of ν;

[0021] S12: Construct a digital twin entity model containing time-varying dynamic terms under complex sea conditions according to the ASV dynamic model, and its model expression is

[0022]

[0023] H t (ν, τ, d) = M -1 (-C(v)v - D(v)v + d) + (M -1 -(M * ) -1 )q

[0024] where: M * represents the nominal mass of the non-diagonal inertial matrix; H t (ν, τ, d) represents the time-varying dynamic terms under complex sea conditions of the ASV dynamic model; M represents the non-diagonal inertial mass matrix;

[0025] S13: According to the digital twin entity model containing time-varying dynamic terms under complex sea conditions, obtain an optimized digital twin entity model containing LSTM modeling errors based on the long short-term memory network LSTM, and its model expression is

[0026]

[0027] where: H0 represents the LSTM modeling error and H0 = [H u0 , H​​v0 ,H r0 T ;H u0 ,H v0 ,H r0 respectively represent the forward velocity error, lateral velocity error, and yaw angular velocity error based on LSTM modeling; represents the learned value of the time-varying dynamic term under complex sea conditions of the ASV dynamic model.

[0028] Further, the S2 specifically includes the following steps:

[0029] S21: According to the optimized digital twin entity model, construct an online learning strategy based on the long short-term memory network LSTM to learn and obtain the time-varying dynamic term in real time;

[0030] And constructing the online learning strategy based on the long short-term memory network LSTM includes an adaptive neuron counting mechanism and a hybrid optimizer switching mechanism;

[0031] The adaptive neuron counting mechanism: Based on the constructed training loss function, dynamically adjust the number of LSTM neurons according to the training loss, and the expression of the adaptive neuron counting mechanism is

[0032]

[0033] In the formula: N represents the number of neurons in the long short-term memory network LSTM; ΔN represents the increase or decrease in the number of neurons in the long short-term memory network LSTM; N min expresses the minimum value of the number of neurons; threshold upper represents the upper bound of the training loss; threshold lower represents the lower bound of the training loss;

[0034] The expression of the constructed training loss function is

[0035]

[0036] In the formula: E(LSTM) represents the expected loss of the online learning strategy based on the long short-term memory network LSTM; represents the learned value of the time-varying dynamic term under complex sea conditions of the ASV dynamic model;

[0037] The hybrid optimizer switching mechanism: Based on the Adam Optimizer, quickly adjust the model weights during the early training period to adapt to the dynamic gradient changes; based on the SGD Optimizer, fine-tune the weight adjustment in the later stage of training to ensure the stability and accuracy of the model;

[0038] And the expression of the hybrid optimizer switching mechanism is​

[0039]

[0040] Where: θ t represents the model parameters of the long short-term memory network LSTM at iteration t; α represents the learning rate; represents the first moment estimate of the time-varying dynamic term bias correction under the Adam Optimizer; represents the second moment estimate of the velocity vector bias correction for adaptive scaling; ε represents the design constant to prevent division by zero during the update based on the Adam Optimizer; gt represents the gradient of the training loss with respect to the model parameters at iteration t; t switch represents the iteration step threshold for the optimizer to switch from the Adam Optimizer to the SGD Optimizer;

[0041] S22: Construct a meta-learning extended state observer, and the meta-learning extended state observer is expressed as

[0042]

[0043] Where: represents the estimate of η; represents the first derivative of; represents the estimate of v; represents the estimate of H0; M1, M2, M3 represent intermediate parameters and I3 represents the 3×3 identity matrix; represents the observer parameters of the meta-learning extended state observer; represents the first derivative of and

[0044] S23: Approximate and process the optimized digital twin entity model according to the meta-learning extended state observer to obtain the estimated value of the LSTM modeling error and the estimated value of the time-varying dynamic term and obtain the ASV dynamics estimation model according to the optimized digital twin entity model after the approximation process combined with the ASV dynamics model.

[0045] Further, the S3 specifically includes the following steps:

[0046] S31: Define the desired trajectory ηι(t) of the ASV and ηι(t) = [xι t , yι t , ψι t T ;

[0047] where, xι t , yι​t , ψi t respectively represent the desired position and the desired heading angle corresponding to the ASV time t;

[0048] Then, the relative distance d and the relative azimuth angle φ of the ASV are obtained according to the ASV dynamics estimation model, and their expressions are

[0049]

[0050]

[0051] S32: Define the ASV tracking position error according to the relative distance d and the relative azimuth angle φ, and its expression is

[0052]

[0053] In the formula: e k represents the ASV tracking position error; represents the actual relative distance or the actual relative azimuth angle; represents the desired relative distance or the desired relative azimuth angle;

[0054] S33: Define the S-shaped function used to not necessarily constrain the initial tracking position error e k (0) of the underactuated ASV system within the predefined region;

[0055] And the expression of the S-shaped function S(t) is

[0056]

[0057] In the formula: α and T q represent the design parameters; t represents the time parameter;

[0058] S34: Based on the S-shaped function S(t), obtain the tracking modulation error E k , and E k = S(t)e k , in order to obtain the initial condition of the tracking position error;

[0059] And the expression of the initial condition of the tracking position error is

[0060]

[0061] In the formula: represent the minimum and maximum values of the relative distance; represents the desired relative distance; represents the desired relative azimuth angle; represents the upper bound of the relative azimuth angle;

[0062] S35: Construct the specified-time performance function, whose expression is

[0063]

[0064] where: α k represents the intermediate parameter piecewise function; t0 represents the starting time, i.e., the moment when the control process begins; ψ k represents the function that changes with time to complete the target task within the set time T f ; represents the steady-state value of z k (t); β k and μ k represent positive design constants; z k (t) represents the error convergence control variable of the specified-time performance function; represents the first derivative of z k (t) and k = d, φ; ψ k represents the intermediate parameter; represents the rate constant for controlling the time decay; T m represents the transition time; T f represents the solidification time; λ k represents the constant gain; T ψ represents the intermediate parameter and T ψ = cos((π(t - T m ))(T f - T m )); -1 );

[0065] S36: Based on the specified-time performance function, obtain the error performance specification constraint according to the initial condition of the tracking position error, and the expression of the error performance specification constraint is

[0066]

[0067] where: δ 1d , δ 2d , δ 1φ , δ 2φ represents the positive proportional coefficient for adjusting the error convergence speed; z d , z φ represents the control variables for adjusting the error convergence of the relative distance and relative azimuth angle.

[0068] S37: Construct the error conversion function θ k according to the error performance specification constraint, and its expression is

[0069]

[0070] And according to the error conversion function θ k, construct a parallel tracking control law for realizing the synchronous mapping of the optimized digital twin entity model and the ASV dynamics model;

[0071] And the expression of the parallel tracking control law is

[0072]

[0073] In the formula: γ = [γ u , γ v , γ r T represents the parallel tracking virtual control law of the forward speed, lateral speed and yaw angular velocity; μ represents the intermediate parameter quantity and μ = [μ u , μ v , μ r , μ u = α u tanh(β u ), μ v = α v tanh(β v ), μ r = α r tanh(β r ); α u , α v , α r represents the design parameters, β u , β v , β v represents the auxiliary system variables; Σ d represents the intermediate parameter quantity; represents the first derivative of; represents as the design parameter and, represents the design parameter, and is obtained from the estimated value l of the forward leading speed u of the ship and the estimated value l of the lateral leading speed v ; represents the estimation of; π k , represents the error transformation scale factor; x φ represents the adjustment function of the yaw angle.

[0074] Furthermore, the S4 specifically includes the following steps:

[0075] S41: Obtain the tracking control error quantity according to the parallel tracking control law;

[0076] And the expression of the tracking variable error is

[0077] ​z = v - γ - μ

[0078] z = [z u ,z v ,z r T ,μ = [μ u ,μ v ,μ r

[0079] S42: Construct an event-triggered controller with a relative threshold according to the tracking variable error and parallel tracking control, and the expression of the event-triggered controller with a relative threshold is

[0080]

[0081] where: ξ p1 ,η p1 represents the gain coefficient or adjustment factor ε for adjusting the dynamic response of the control input p represents the error term related to the control signal; represents the control parameter for obtaining the time-varying dynamic term learning value of the ASV dynamics model under complex sea conditions based on the meta-learning extended state observer; μ p represents the system adjustment parameter for controlling the dynamic response of the error, represents the reference mass related to the control inputs u and r; represents the estimated parameter for adjusting the control input; ε c represents the adjustment constant; k p ,Γ p ,v p represents positive design parameters; represents the first derivative of; ω p and ω v represents the intermediate parameter quantity and ω u = θ d π d cosφ, ω v = -θ d π d sinv and ω r = z φ π φ ; τ p represents the system control input;

[0082] S43: Design a memory-based event-triggered mechanism for dynamically adjusting the event-triggering condition to optimize communication resources based on the event-triggered controller with a relative threshold, so as to achieve the performance control of the underactuated autonomous surface vehicle at a predetermined time;

[0083] And the designed memory-based event-triggered mechanism, its expression is ​​

[0084]

[0085] |e p (t)|≥N + η p3

[0086] N = ξ p1 (|η p1 q p (t)+η p2 q p (t - τ)|)

[0087] η p1 (t)=1 - η p2 (t),

[0088]

[0089] In the formula: represents the output of the event - triggered controller at time t; e j (t) represents the triggering error of the memory - based event - triggered mechanism and p (t) represents the actual control input within the time interval [t ξ p1 , ξ p2 , represents a preset threshold parameter; η p1 , η p2 , η p3 represents an intermediate parameter quantity; q p (t) represents the actual control input within the time interval [t j , t j+1 ) and t j , t j+1 represents a time parameter; q p (t - τ) represents the control input at the delayed time τ; η p2 (t - τ) represents the intermediate parameter quantity at the delayed time τ; N represents an intermediate parameter quantity; represents the first - order derivative of η p3 with respect to.

[0090] The present invention provides a predetermined - time performance control method for an under - actuated autonomous surface vehicle based on digital twin, and the beneficial effects are as follows:

[0091] (1) By constructing a digital twin entity model containing unknown time - varying dynamic terms under complex sea conditions, and using the long short - term memory network LSTM to obtain an optimized digital twin entity model containing LSTM modeling errors, the present invention solves the requirement for large - scale data storage in traditional trajectory tracking tasks, provides real - time estimation and compensation for LSTM modeling errors, and enhances interpretability;

[0092] (2) Through the constructed online learning strategy based on the long short-term memory network (LSTM), combined with the constructed meta-learning extended state observer, estimate the LSTM modeling error to update the digital optimized twin entity model, and design a parallel tracking control law for realizing the synchronous mapping of the optimized digital twin entity model and the ASV dynamics model through the constructed specified-time performance function; by introducing the specified-time performance function, the user is allowed to customize the convergence time and eliminate the constraint that the initial error must be within the performance boundary. Through the designed parallel tracking control law, the system synchronization between the digital twin entity model DTE and the actual ASV dynamics model ASV is achieved, realizing the real-time mapping between the virtual space and the physical space.

[0093] (3) Construct an event-triggered controller with a relative threshold according to the parallel tracking control law, and then design a memory-based event-triggered mechanism for dynamically adjusting the event-triggering condition to optimize communication resources, greatly optimizing the use of communication resources, enabling high control accuracy and stability under complex sea conditions, and effectively reducing the consumption of communication resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] 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-described drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0095] Figure 1 It is a flowchart of the predetermined-time performance control method for an underactuated autonomous surface vehicle based on digital twin of the present invention;

[0096] Figure 2 It is a schematic diagram of the position error in this embodiment;

[0097] Figure 3 It is a schematic diagram of the angle error in this embodiment;

[0098] Figure 4 It is a schematic diagram of the learning performance of SLTM in this embodiment;

[0099] Figure 5 It is a schematic diagram of the observation accuracy in this embodiment;

[0100] Figure 6 It is a schematic diagram of the control input in this embodiment;

[0101] Figure 7 It is a schematic diagram of the time interval between events in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0102] 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.

[0103] This embodiment provides a predetermined time performance control method for an underactuated autonomous surface vehicle based on digital twin, as Figure 1 shown, which specifically includes the following steps:

[0104] S1: Establish an ASV dynamic model of the underactuated autonomous surface vehicle, and construct a digital twin entity model including unknown time-varying dynamic terms under complex sea conditions according to the ASV dynamic model; and obtain an optimized digital twin entity model including LSTM modeling errors based on the long short-term memory network LSTM;

[0105] Specifically, it includes the following steps:

[0106] S11: Establish an ASV dynamic model of the underactuated autonomous surface vehicle, and its model expression is

[0107]

[0108] where: η represents the position and heading angle of the ASV and η = [x, y, ψ] T ; v represents the velocity vector and v = [u, v, r] T ; u, v, r represent the forward velocity, lateral velocity, and yaw angular velocity; q represents the control input and τ = q = [q u , 0, q r T ; q u , q r represent the control inputs provided by the thrusters and rudders; M, C(v), D(v) represent the mass matrix, Coriolis matrix, and hydrodynamic damping matrix respectively; d represents the unknown disturbance and d = [d u , d v , d r T ; d u , d v , d r represent the external unknown disturbances of the underactuated ASV; R(ψ) represents the rotation matrix used to convert the hull coordinate system to the inertial coordinate system; represents the first derivative of η; represents the first derivative of v;

[0109] S12: Decompose the ASV dynamics model into a known nominal part and an unknown time-varying dynamic term, and use a long short-term memory network (LSTM) to learn the unknown time-varying dynamic term online to construct a digital twin entity model that includes the time-varying dynamic term under complex sea conditions. Its model expression is

[0110]

[0111] H t (v,τ,d) = M -1 (-C(v)v - D(v)v + d)+(M -1 -(M * ) -1 )q

[0112] In the formula: M * represents the nominal parameter of the non-diagonal inertia matrix M; H t (v,τ,d) represents the time-varying dynamic term under complex sea conditions of the ASV dynamics model; M represents the non-diagonal inertia mass matrix;

[0113] S13: In this embodiment, to solve the problem of unknown time-varying dynamic terms under complex sea conditions, a method based on LSTM is applied to learn the unknown ASV time-varying dynamic term online, and the learned ASV time-varying dynamic term is expressed as Therefore, according to the digital twin entity model that includes the time-varying dynamic term under complex sea conditions, an optimized digital twin entity model including the LSTM modeling error is obtained based on the long short-term memory network LSTM, that is, the digital twin entity model is re-expressed as

[0114]

[0115] In the formula: H0 represents the LSTM modeling error and H0 = [H u0 ,H v0 ,H r0 T ; H u0 ,H v0 ,H r0 respectively represent the forward velocity error, lateral velocity error, and yaw angular velocity error based on LSTM modeling; represents the learned value of the time-varying dynamic term under complex sea conditions of the ASV dynamics model;

[0116] S2: According to the optimized digital twin entity model, construct an online learning strategy based on the long short-term memory network LSTM to learn and obtain the time-varying dynamic term in real time; and combine the constructed meta-learning extended state observer to estimate the LSTM modeling error and the time-varying dynamic term to obtain the ASV dynamics estimation model;

[0117] Specifically, it includes the following steps:​

[0118] S21: Construct an online learning strategy based on the long short-term memory network (LSTM) according to the optimized digital twin entity model to learn and obtain time-varying dynamic terms in real time;

[0119] And the construction of the online learning strategy based on the long short-term memory network (LSTM) includes an adaptive neuron counting mechanism and a hybrid optimizer switching mechanism;

[0120] In this embodiment, to improve the flexibility and efficiency of the model, the number of LSTM neurons is dynamically adjusted according to the training loss. When the loss exceeds the set threshold, the number of neurons N increases to improve the learning ability; on the contrary, when the loss remains low, the number of neurons N decreases to prevent overfitting and reduce the computational cost;

[0121] And the adaptive neuron counting mechanism: Based on the constructed training loss function, dynamically adjust the number of LSTM neurons according to the training loss, and the expression of the adaptive neuron counting mechanism is

[0122]

[0123] In the formula: N represents the number of neurons in the long short-term memory network (LSTM); ΔN represents the increase or decrease in the number of neurons in the long short-term memory network (LSTM); N min expresses the minimum value of the number of neurons; threshold upper represents the upper bound of the training loss; threshold lower represents the lower bound of the training loss;

[0124] The expression of the constructed training loss function is

[0125]

[0126] In the formula: E(LSTM) represents the expected loss of the online learning strategy based on the long short-term memory network (LSTM); represents the learning value of the time-varying dynamic term under complex sea conditions of the ASV dynamics model;

[0127] In this embodiment, through the adaptive neuron counting mechanism, the model can be allowed to expand its complexity when necessary and simplify when possible, thereby optimizing computational resources;

[0128] In order to further improve the training efficiency and stability, this embodiment constructs a hybrid optimizer switching mechanism, and the hybrid optimizer switching mechanism: rapidly adjusts the model weights based on Adam Optimizer during the early training to adapt to dynamic gradient changes; fine-tunes the weight adjustment based on SGD Optimizer in the later stage of training to ensure the stability and accuracy of the model; where the early training and the later stage of training are divided into the early training and the later stage of training according to the total number of iterations set, and a threshold of the number of iterations is preset according to empirical values, so as to divide the entire training stage into the early training and the later stage of training according to the preset threshold of the number of iterations;

[0129] And the expression of the hybrid optimizer switching mechanism is

[0130]

[0131] In the formula: θ t represents the model parameters of the long short-term memory network LSTM at iteration t; α represents the learning rate; represents the first-order moment estimation of the time-varying dynamic term bias correction based on Adam Optimizer, and its function is the momentum term; represents the second-order moment estimation of the velocity vector bias correction for adaptive scaling; ε represents the design constant to prevent division by zero during the update based on Adam Optimizer; gt represents the gradient of the training loss with respect to the model parameters at iteration t; t switch represents the iteration step threshold for the optimizer to switch from Adam Optimizer to SGD Optimizer;

[0132] In this embodiment, Adam optimizer accelerates convergence in the early stage, and SGD optimizer ensures stable convergence in the later stage, greatly improving the flexibility of model training, training efficiency, and the stability of the trained model; among them, the implementation process of the online learning strategy based on the long short-term memory network LSTM in this embodiment is specifically as follows: online modeling of the digital twin entity model based on the long short-term memory network LSTM:

[0133] Input: control input q, velocity ν, disturbance d;

[0134] Output: estimated model uncertainty, that is, unknown time-varying dynamic term

[0135] Initialize the long short-term memory network LSTM model with the initial neuron count N0;

[0136] Set the learning rate α and the threshold threshold for the loss upper and threshold lower ;

[0137] Set the switching point, i.e., the iteration step threshold t switch For switching from the Adam optimizer to the SGD;

[0138] For k = 1, 2, …, T, collect the dynamic state x of the current ASV k (η and ν), the control input u k (τ) and the time step t k As the model input for initializing the long short - term memory network LSTM;

[0139] The forward - propagation formula of the long short - term memory network LSTM is as follows:

[0140] Calculate the forget gate: H t = σ(W f ·[h t-1 , x t +b f );

[0141] Calculate the input gate: i t = σ(W i ·[h t-1 , x t +b i );

[0142] Calculate the candidate memory:

[0143] Update the cell state:

[0144] Calculate the output gate: o t = σ(W o ·[h t-1 , x t +b o );

[0145] Update the hidden state: h t = o t ·tanh(C t );

[0146] Predict the uncertainty in the ASV dynamic model, i.e., the unknown time - varying dynamic term Obtain the actual value of the uncertainty H t (v, τ, d) from the feedback of the existing ASV dynamic model system; and calculate the loss E(LSTM) according to the predicted value of the time - varying dynamic term and the actual value H t (ν, τ, d); if L k > threshold upper , then increase the neuron count N = N + ΔN and re - initialize the LSTM model; otherwise if Lk <threshold lower , then reduce the neuron count N = N - ΔN, and re-initialize the model parameters of the long short-term memory network LSTM; if K < t switch , then end updating the model parameters of the long short-term memory network LSTM using the Adam optimizer; otherwise, update the model parameters of the long short-term memory network LSTM using the SGD optimizer;

[0147] In this embodiment, the LSTM adjusts its weights through backpropagation to minimize this loss. In the time-dependent ASV dynamics model, the LSTM network is used to effectively approximate the uncertain dynamics H(x k , u k , t k ); through the storage unit and gating mechanism of the LSTM, it can capture complex temporal relationships and estimate complex non-linear dynamics. In this embodiment, to improve the model performance and efficiency, based on the adaptive neuron mechanism, the network complexity is dynamically adjusted according to the task, and through the hybrid optimizer strategy (combining Adam and SGD), fast convergence during the early training period and stability in the later stage are ensured. This combination improves the accuracy of approximating uncertainty and optimizes the use of computing resources;

[0148] S22: Construct a meta-learning extended state observer, and the representation of the meta-learning extended state observer is

[0149]

[0150] where: represents the estimate of η; represents the first derivative of; represents the estimate of v; represents the estimate of H0; M1, M2, M3 represent intermediate parameters and I3 represents a 3×3 identity matrix; represents the observer parameters of the meta-learning extended state observer; represents the first derivative of and

[0151] S23: Approximate and process the digital twin entity model according to the meta-learning extended state observer to obtain the estimated value of the LSTM modeling error and the estimated value of the time-varying dynamic term and obtain the ASV dynamics estimation model according to the optimized digital twin entity model after the approximation process combined with the ASV dynamics model.

[0152] S3: Based on the updated optimized digital twin entity model and the ASV dynamics model, design a parallel tracking control law for realizing the synchronous mapping of the optimized digital twin entity model and the ASV dynamics model based on the constructed performance function within a specified time; the role of the meta-learning extended state observer constructed in S22 of this embodiment is to synchronize the actual ASV state with the ASV state in the virtual space. As long as the LSTM modeling error is small enough, parallel control between the virtual and the actual can be achieved, which is the parallel tracking control law;

[0153] Specifically, it includes the following steps:

[0154] S31: Define the desired trajectory ηι(t) of the ASV, and ηι(t) = [xι t , yι t , ψι t T ;

[0155] Among them, xι t , yι t , ψι t respectively represent the desired position and the desired heading angle of the ASV at time t;

[0156] Then, obtain the relative distance d and the relative azimuth angle φ of the ASV according to the ASV dynamics estimation model, and their expression is

[0157]

[0158]

[0159] S32: Define the ASV tracking position error according to the relative distance d and the relative azimuth angle φ, and its expression is

[0160]

[0161] In the formula: e k represents the ASV tracking position error; represents the actual relative distance or the actual relative azimuth angle; represents the desired relative distance or the desired relative azimuth angle;

[0162] S33: Define an S-shaped function for initially tracking the position error e k (0) of the underactuated ASV system, which does not necessarily need to be constrained within a predefined region;

[0163] And the expression of the S-shaped function S(t) is

[0164]

[0165] In the formula: α and T q ​represents design parameters; t represents time parameters;

[0166] S34: Based on the S-shaped function S(t), obtain the tracking modulation error E according to the tracking position error k , and E k = S(t)e k , to obtain the initial condition of the tracking position error;

[0167] And the expression of the initial condition of the tracking position error is

[0168]

[0169] In the formula: represents the minimum and maximum values of the relative distance; represents the desired relative distance; represents the desired relative azimuth angle; represents the upper bound of the relative azimuth angle;

[0170] S35: Construct a specified-time performance function, and its expression is

[0171]

[0172]

[0173] In the formula: α k represents the intermediate parameter piecewise function; t0 represents the starting time, that is, the moment when the control process starts; ψ k represents the function that changes with time to complete the target task within the set time T f ; represents z k (t)'s steady-state value; β k and μ k represent positive design constants; z k (t) represents the error convergence control variable of the specified-time performance function; represents z k (t)'s first derivative and k = d, φ; ψ k represents the intermediate parameter; represents the rate constant for controlling time decay; T m represents the transition time; T f represents the solidification time; λ k represents the constant gain; T ψ represents the intermediate parameter and T ψ = cos((π(t - T m ))(T f - T m ) -1 );

[0174] S36: To ensure high performance in the trajectory tracking task, this embodiment imposes the following performance specifications on the tracking position error, that is, based on the specified time performance function, the error performance specification constraint is obtained according to the initial conditions of the tracking position error, and the expression of the error performance specification constraint is

[0175]

[0176] where: δ 1d , δ 2d , δ 1φ , δ 2φ represents a positive proportionality coefficient used to adjust the error convergence rate; z d , z φ represents a control variable used to adjust the convergence of the relative distance and relative azimuth angle errors.

[0177] S37: An error conversion function θ k is constructed according to the error performance specification constraint, and its expression is

[0178]

[0179] And a parallel tracking control law for realizing the synchronous mapping of the optimized digital twin entity model and the ASV dynamics model is constructed according to the error conversion function θ k .

[0180] And the expression of the parallel tracking control law is

[0181]

[0182]

[0183] where: γ = [γ u , γ v , γ r T represents the parallel tracking virtual control law of the forward speed, lateral speed, and yaw angular velocity; μ represents an intermediate parameter and μ = [μ u , μ v , μ r , μ u = α u tanh(β u ), μ v = α v tanh(β v ), μ r = α r tanh(β r ); α u , α v , α r represents design parameters, β u , β​v , β v represents an auxiliary system variable; Σ d represents an intermediate parameter quantity; represents the first derivative of; is represented as a design parameter and, is a design parameter and is obtained from the estimated value of the forward leading speed u of the ship l the estimated value of and the estimated value of the lateral leading speed v l the estimated value of obtained; represents the estimation of; π k , represents the error transformation scale factor; x φ represents the adjustment function of the yaw angle;

[0184] In this embodiment, by designing a parallel tracking control law, the system of the actual ASV dynamics model can be smoothly transitioned in parallel with its twin entity model DTE, realizing the synchronous mapping between the physical space and the cyber space. The twin entity model DTE is adaptively adjusted through the online learning ability of LSTM to respond to the changing environment, enabling the ASV dynamics model to gradually optimize its behavior and prediction ability in a dynamic environment;

[0185] S4: Construct an event-triggered controller with a relative threshold according to the parallel tracking control law, and design a memory-based event-triggered mechanism for dynamically adjusting the event-triggering condition to optimize communication resources based on the event-triggered controller with a relative threshold, so as to achieve the performance control of the underactuated autonomous surface vehicle at a predetermined time;

[0186] Specifically, it includes the following steps:

[0187] S41: Obtain the tracking control error quantity according to the parallel tracking control law;

[0188] And the expression of the tracking variable error is

[0189] z = ν - γ - μ

[0190] z = [z u , z v , z r T , μ = [μ u , μ v , μ r

[0191] S42: Construct an event-triggered controller with a relative threshold according to the tracking variable error and the parallel tracking control, and the expression of the event-triggered controller with a relative threshold is​​

[0192]

[0193] Among them, the parameter update law and the auxiliary system of the event-triggered controller with a relative threshold are expressed as

[0194]

[0195] In the formula: ξ p1 , η p1 denote the gain coefficient or the adjustment factor ε for adjusting the dynamic response of the control input p denotes the error term related to the control signal; denotes the control parameter for obtaining the time-varying dynamic term learning value of the ASV dynamics model under complex sea conditions based on the meta-learning extended state observer; μ p denotes the system adjustment parameter for controlling the dynamic response of the error, denotes the reference mass related to the control inputs u and r; denotes the estimated parameter for adjusting the control input; ε c denotes the adjustment constant; k p , Γ p , v p denote positive design parameters; denotes the first derivative of; ω p related to ω v denotes the intermediate parameter quantity and ω u = θ d π d cosφ, ω v = -θ d π d sinφ and ω r = z φ π φ ; τ p denotes the system control input;

[0196] S43: Design a memory-based event-triggered mechanism based on an event-triggered controller with a relative threshold for dynamically adjusting the event-triggering condition to optimize communication resources, so as to achieve the performance control of the underactuated autonomous surface vehicle at a predetermined time;

[0197] And the designed memory-based event-triggered mechanism, whose expression is

[0198]

[0199] |e p (t)| ≥ N + η p3

[0200] N = ξ p1(|η p1 q p (t)+η p2 q p (t - τ)|)

[0201] η p1 (t)=1 - η p2 (t),

[0202]

[0203] In the formula: represents the output of the event - triggered controller at time t; e j (t) represents the triggering error of the memory - based event - triggering mechanism and p ξ ξ p1 , ξ p2 , represents the preset threshold parameter; η p1 , η p2 , η p3 represents the intermediate parameter quantity; q p (t) represents the actual control input within the time interval [t j , t j+1 ) ; t j , t j+1 represents the time parameter; q p (t - τ) represents the control input at the delayed time τ; η p2 (t - τ) represents the intermediate parameter quantity at the delayed time τ; N represents the intermediate parameter quantity; represents the first - order derivative of η p3 .

[0204] This embodiment also includes the stability proof process of this method:

[0205] Proof: Define the error variable as

[0206] Let The following formula can be obtained:

[0207]

[0208] Among them, the matrices C and D are defined as:

[0209]

[0210] There exists a matrix P such that:

[0211] PC + C T P+βI6=0,

[0212] Where is a positive constant. Consider the Lyapunov function V2:

[0213]

[0214] Taking its derivative along the system trajectory, in this embodiment, we have

[0215]

[0216] Considering the properties of P, in this embodiment, we obtain:

[0217]

[0218] Now, by restricting the term, in this embodiment, we get

[0219]

[0220] where represents a constant related to the external input ; if the external input is bounded, then γ is also a bounded constant. Based on the standard results in Lyapunov function analysis, in this embodiment, it can be concluded that the ASV system state S is SGUUB.

[0221] This embodiment also includes an experimental example to verify the effectiveness of the method:

[0222] The reference trajectory in this embodiment is designed as follows:

[0223]

[0224] where

[0225] ΔN = 5threshold upper = 0.1threshold lower = 0.01

[0226] N min = 10t switch = 20M1 = [10, 20, 10] T

[0227] M2 = [15, 21, 11] T M3 = [31, 33, 24] T

[0228] ξ u1 = 7ξ u2 = 6

[0229] ξ r1 = 3ξ r2 = 1.5

[0230] τ = 1

[0231]

[0232] α u = 30

[0233] α v = 20α r = 15

[0234] κ u = 5κ v = 3

[0235] κ r = 2u l = 3v l = 0

[0236] In this embodiment, the initial conditions of the leader are defined as η l (0) = [0, 0, 0] T ; the initial conditions of the follower are defined as η = [-14, -13, π / 6] T and v = [0, 0, 0] T ; the data of the ASV is collected online for the purpose of real-time training of the LSTM. The LSTM model starts with 70 neurons and accepts 11-dimensional data (8 input features and 3 target outputs); it uses the Adam optimizer (for the first 20 steps, the learning rate is 0.0001, the momentum coefficient is 0.95, 0.99) for weight update, and then switches to the SGD optimizer; in order to further verify the comparison results, common performance metrics are introduced for quantitative analysis, including the mean absolute error and the mean value. The experimental results are as Figures 2 to 7 shown; as Figures 2 to 3 is the position and heading angle error curve, indicating that the PTPF function converging based on user-defined time makes the error quickly converge to the boundary within the preset time without initial error constraint; Figure 4 the loss function curve of the LSTM in Figure 5 verifies the effectiveness of the online learning mechanism; Figures 6 to 7 shows the speed and modeling error of the meta-learning state observer based on the LSTM; as

[0237] The beneficial effects of the method in this embodiment are as follows:

[0238] (1) The present invention constructs a digital twin entity model containing unknown time-varying dynamic terms under complex sea conditions, and uses the long short-term memory network LSTM to obtain an optimized digital twin entity model containing LSTM modeling errors, solving the problems of the lack of virtual reality interaction due to the need for large-scale data storage in the trajectory tracking task, providing real-time estimation and compensation of LSTM modeling errors, and enhancing interpretability;

[0239] (2) Through the constructed online learning strategy based on the long short-term memory network LSTM, combined with the constructed meta-learning extended state observer, the LSTM modeling error is estimated to update the digital optimized twin entity model, and a parallel tracking control law for realizing the synchronous mapping of the optimized digital twin entity model and the ASV dynamics model is designed based on the constructed specified-time performance function; by introducing the specified-time performance function, the user is allowed to customize the convergence time and eliminate the constraint that the initial error must be within the performance boundary, and completely eliminates the limitation in the traditional preset performance control PPC that the initial error must be located within the preset boundary. Even if the initial error exceeds the performance boundary, it can still be mapped to the controllable range through the error conversion function, ensuring that the tracking error converges stably to the steady-state value within the specified time. This feature significantly improves the robustness and adaptability of the underactuated ASV trajectory tracking in complex marine environments. In addition, through the designed parallel tracking control law, the system synchronization of the digital twin entity model DTE and the actual ASV dynamics model ASV is realized, achieving real-time mapping between the virtual space and the physical space;

[0240] (3) According to the parallel tracking control law, an event-triggered controller with a relative threshold is constructed, and then a memory-based event-triggered mechanism for dynamically adjusting the event-triggering condition to optimize communication resources is designed. By introducing a memory factor, the weights of historical data and real-time data are dynamically balanced to optimize the triggering condition; compared with the traditional event-triggered mechanism, this mechanism reduces the number of communication trigger times by 45%, effectively reducing bandwidth occupancy and energy consumption, while suppressing the false triggering problem caused by instantaneous data fluctuations, and is applicable to scenarios with limited marine communication resources, greatly optimizing the use of communication resources, enabling high control accuracy and stability under complex sea conditions, and being able to effectively reduce the consumption of communication resources.

[0241] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; 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 make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A predetermined-time performance control method for an underactuated autonomous surface vehicle based on digital twin, characterized in that Specifically, it includes the following steps: S1: Establish the ASV dynamic model of the underactuated autonomous surface vehicle, and construct a digital twin entity model containing unknown time-varying dynamic terms under complex sea conditions according to the ASV dynamic model; And based on the long short-term memory network LSTM, obtain an optimized digital twin entity model containing LSTM modeling errors; S2: According to the optimized digital twin entity model, construct an online learning strategy based on the long short-term memory network LSTM to learn and obtain time-varying dynamic terms in real time; And combine the constructed meta-learning extended state observer to estimate the LSTM modeling error and time-varying dynamic terms to obtain the ASV dynamic estimation model; S3: According to the ASV dynamic estimation model, design a parallel tracking control law for realizing the synchronous mapping of the optimized digital twin entity model and the ASV dynamic model based on the constructed specified-time performance function; S4: Construct an event-triggered controller with a relative threshold according to the parallel tracking control law; And based on the event-triggered controller with a relative threshold, design a memory-based event-triggered mechanism for dynamically adjusting the event-triggering condition to optimize communication resources to achieve the performance control of the underactuated autonomous surface vehicle at a predetermined time.

2. The predetermined time performance control method of an underactuated autonomous surface vehicle based on digital twin according to claim 1, characterized in that, The specific steps of S1 are as follows: S11: Establish the ASV dynamic model of the underactuated autonomous surface vehicle, and its model expression is Where: η represents the position and heading angle of the ASV, and η = [x, y, ψ] T ; v represents the velocity vector, and v = [u, v, r] T ; u, v, r represent the forward velocity, lateral velocity, and yaw angular velocity; q represents the control input, and τ = q = [q u , 0, q r T ; q u , q r represents the control input provided by the thrusters and rudder; M, C(v), D(ν) represent the mass matrix, Coriolis matrix, and hydrodynamic damping matrix, respectively; d represents the unknown disturbance, and d = [d u , d v , d r T ; d u , d v , d r represents the external unknown disturbance of the underactuated ASV; R(ψ) represents the rotation matrix used to transform the hull coordinate system into the inertial coordinate system; represents the first derivative of η; represents the first derivative of v;​​ S12: According to the ASV dynamic model, construct a digital twin entity model containing time-varying dynamic terms under complex sea conditions, and its model expression is H t (v,τ,d) = M -1 (-C(v)v - D(v)v + d) + (M -1 -(M*) -1 )q Where: M * represents the nominal mass of the non - diagonal inertia matrix; H t (v, τ, d) represents the time - varying dynamic term under complex sea conditions of the ASV dynamics model; M represents the non - diagonal inertia mass matrix; S13: According to the digital twin entity model containing time-varying dynamic terms under complex sea conditions, obtain an optimized digital twin entity model containing LSTM modeling errors based on the long short-term memory network LSTM, and its model expression is Where: H0 represents the LSTM modeling error and H0 = [H u0 , H v0 , H r0 T ; H u0 , H v0 , H r0 represent the forward velocity error, lateral velocity error, and yaw angular velocity error based on LSTM modeling, respectively; represents the learned value of the time-varying dynamic term under complex sea conditions of the ASV dynamics model.​ 3. A predetermined-time performance control method for an underactuated autonomous surface vehicle based on digital twin according to claim 2, wherein The specific steps of S2 are as follows: S21: According to the optimized digital twin entity model, construct an online learning strategy based on the long short-term memory network LSTM to learn and obtain time-varying dynamic terms in real time; And the constructed online learning strategy based on the long short-term memory network LSTM includes an adaptive neuron counting mechanism and a hybrid optimizer switching mechanism; The adaptive neuron counting mechanism: Based on the constructed training loss function, dynamically adjust the number of LSTM neurons according to the training loss, and the expression of the adaptive neuron counting mechanism is Where: N represents the number of neurons in the long short-term memory network LSTM; ΔN represents the increase or decrease in the number of neurons in the long short-term memory network LSTM; N min represents the minimum value of the number of neurons; threshold upper represents the upper bound of the training loss; threshold lower represents the lower bound of the training loss; The expression of the constructed training loss function is where: E(LSTM) represents the expected loss of the online learning strategy based on the long short-term memory network LSTM; represents the learning value of the time-varying dynamic term under complex sea conditions of the ASV dynamics model; The hybrid optimizer switching mechanism: Based on the Adam Optimizer, quickly adjust the model weights during the early training period to adapt to dynamic gradient changes; Based on the SGD Optimizer, fine-tune the weight adjustment in the later stage of training to ensure the stability and accuracy of the model; And the expression of the hybrid optimizer switching mechanism is Where: θ t represents the model parameters of the long short-term memory network LSTM at iteration t; α represents the learning rate; represents the first-order moment estimate of the time-varying dynamic term bias correction under AdamOptimizer; represents the second-order moment estimate of the velocity vector bias correction for adaptive scaling; ε represents the design constant to prevent division by zero during the update based on Adam Optimizer; gt represents the gradient of the training loss with respect to the model parameters at iteration t; t switch represents the iteration step threshold for the optimizer to switch from Adam Optimizer to SGD Optimizer; S22: Construct a meta-learning extended state observer, and the representation of the meta-learning extended state observer is Wherein: represents the estimate of η; represents the first derivative of; represents the estimate of v; represents the estimate of H0; M1, M2, M3 represent intermediate parameters and I3 represents a 3×3 identity matrix; represents the observer parameters of the meta-learning extended state observer; represents the first derivative of and S23: Optimize the digital twin entity model according to the approximation processing of the meta-learning extended state observer to obtain the estimated value of the LSTM modeling error and the estimated value of the time-varying dynamic term And obtain the ASV dynamics estimation model according to the optimized digital twin entity model after the approximation processing and the ASV dynamics model 4. A predetermined time performance control method for an underactuated autonomous surface vehicle based on digital twin according to claim 3, characterized in that The specific steps of S3 are as follows: S31: Define the desired trajectory ηι(t) of the ASV, and ηι(t) = [xι t , yι t , ψι t T ;​ where, xι t , yι t , ψι t respectively represent the desired position and the desired heading angle corresponding to the ASV time t; Then obtain the relative distance d and relative azimuth angle φ of the ASV according to the ASV dynamic estimation model, and its expression is S32: Define the ASV tracking position error according to the relative distance d and relative azimuth angle φ, and its expression is where: e k represents the ASV tracking position error; represents the actual relative distance or the actual relative azimuth angle; represents the desired relative distance or the desired relative azimuth angle; S33: Define an S-shaped function for the initial tracking position error e k (0) of the underactuated ASV system, which does not have to be constrained within a predefined region; And the expression of the S-shaped function S(t) is Where: α and T q represent design parameters; t represents a time parameter; S34: Based on the S-shaped function S(t), obtain the tracking modulation error E according to the tracking position error k , and E k = S(t)e k , to obtain the initial condition of the tracking position error; And the expression of the initial condition of the tracking position error is In the formula: represents the minimum and maximum values of the relative distance; represents the expected relative distance; represents the expected relative azimuth angle; represents the upper bound of the relative azimuth angle; S35: Construct the specified-time performance function, whose expression is Where: α k represents the intermediate parameter piecewise function; t0 represents the starting time, i.e., the moment when the control process begins; ψ k represents the function that changes with time to complete the target task within the set time T f ; represents the steady-state value of z k (t); β k and μ k represent positive design constants; z k (t) represents the error convergence control variable of the specified time performance function; represents the first derivative of z k (t) and k = d, φ; ψ k represents the intermediate parameter; represents the rate constant for controlling the time decay; T m represents the transition time; T f represents the solidification time; λ k represents the constant gain; T ψ represents the intermediate parameter and T ψ = cos((π(t - T m ))(T f - T m )) -1 ); S36: Based on the specified-time performance function, obtain the error performance specification constraint according to the initial condition of the tracking position error, and the expression of the error performance specification constraint is Where: δ 1d , δ 2d , δ 1φ , δ 2φ represents a positive proportionality coefficient for adjusting the error convergence rate; z d , z φ represents a control variable for adjusting the convergence of the relative distance and relative azimuth angle errors; S37: Construct an error conversion function θ according to the error performance specification constraints k , and its expression is and according to the error conversion function θ k , a parallel tracking control law for realizing the synchronous mapping of the optimized digital twin entity model and the ASV dynamics model is constructed; And the expression of the parallel tracking control law is where: γ = [γ u , γ v , γ r T represents the parallel tracking virtual control law for the forward speed, lateral speed, and yaw angular velocity; μ represents the intermediate parameter and μ = [μ u , μ v , μ r , μ u = α u tanh(β u ), μ v = α v tanh(β v ), μ r = α r tanh(β r ); α u , α v , α r represent the design parameters, β u , β v , β v represent the auxiliary system variables; Σ d represents the intermediate parameter; represents the first derivative of; is represented as a design parameter and, represents the design parameter, and is obtained from the estimated values l of the forward pilot speed u and the estimated values l of the lateral pilot speed v ; represents the estimation of ; π k , represents the error transformation scale factor; x φ represents the adjustment function of the yaw angle.​ 5. A predetermined time performance control method for an underactuated autonomous surface vehicle based on digital twin according to claim 4, characterized in that The specific steps of S4 are as follows: S41: Obtain the tracking control error quantity according to the parallel tracking control law; And the expression of the tracking variable error is z = v - γ - μ z = [z u , z v , z r T , μ = [μ u , μ v , μ r ​​ S42: Based on the tracking variable error and the parallel tracking control, construct an event-triggered controller with a relative threshold, and the expression of the event-triggered controller with a relative threshold is where: ξ p1 , η p1 represent the gain coefficient or adjustment factor ε for adjusting the dynamic response of the control input p represents the error term related to the control signal; represents the control parameter for obtaining the time-varying dynamic term learning value of the ASV dynamics model under complex sea conditions based on the meta-learning extended state observer; μ p represents the system adjustment parameter for controlling the dynamic response of the error, represents the reference mass related to the control inputs u and r; represents the estimated parameter for adjusting the control input; ε c represents the adjustment constant; k p , Γ p , ν p represent positive design parameters; represents the first derivative of; ω p with ω v represents an intermediate parameter quantity and ω u = θ d π d cosφ, ω v = -θ d π d sinφ and ω r = z φ π φ ; τ p represents the system control input; S43: Based on the event-triggered controller with a relative threshold, design a memory-based event-triggered mechanism for dynamically adjusting the event-triggering condition to optimize communication resources, so as to achieve the performance control of the underactuated autonomous surface vehicle at the specified time; And the designed memory-based event-triggered mechanism, its expression is |e p (t)|≥N + η p3 N = ξ p1 (|η p1 q p (t) + η p2 q p (t - τ)|) η p1 (t) = 1 - η p2 (t), Wherein: represents the output of the event-triggered controller at time t; e j p (t) represents the triggering error of the memory-based event-triggering mechanism and ξ p1 , ξ p2 , represents the preset threshold parameter; η p1 , η p2 , η p3 represents the intermediate parameter quantity; q p (t) represents the actual control input within the time interval [t j , t j+1 ); t j , t j+1 represents the time parameter; q p (t - τ) represents the control input at the delayed time τ; η p2 (t - τ) represents the intermediate parameter quantity at the delayed time τ; N represents the intermediate parameter quantity; represents the first derivative of η p3 .​

Images