Unmanned ship cluster cooperative tracking control method based on model predictive control and control law learning

Abstract

The invention relates to an unmanned ship cluster cooperative tracking control method based on model predictive control and control law learning, and belongs to the field of unmanned ship formation trajectory tracking control. The method aims at solving the problems that an existing centralized control algorithm is high in calculation complexity, and a distributed control algorithm is poor in control effect. Firstly, kinematics and dynamics models of the unmanned ship are established, then a distributed dual-mode control algorithm and a centralized model predictive control algorithm are designed, a long short-term memory (LSTM) network is constructed to learn a deep mapping relation between a global control law and a distributed control law, and then, the deep mapping relation between the global control law and the distributed control law is learned by aiming at the problem that an offline learning method is poor in expandability. A self-adaptive model switching strategy is designed, the LSTM model is switched in a self-adaptive mode according to the characteristics of the tracking trajectory, the cooperative control effect is improved while the calculation time is shortened, and it is ensured that the multiple unmanned ship systems can track the time-varying target trajectory in real time according to various expected formations.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned boat formation trajectory tracking control, and specifically relates to an unmanned boat cluster collaborative tracking control method based on model predictive control and control law learning. Background Art

[0002] An unmanned surface vehicle (USV) is a surface vehicle platform with environmental perception, intelligent decision-making and autonomous navigation capabilities. Thanks to its endurance and load capacity, unmanned ships play an important role in water tasks such as environmental monitoring, marine rescue and hydrological exploration. Compared with a single unmanned ship, a swarm of unmanned ships can perform complex water tasks more flexibly. Therefore, research related to swarms of unmanned ships has received increasing attention in recent years. As one of the core issues of unmanned ship swarm control, the collaborative trajectory tracking task requires the swarm of unmanned ships to maintain a certain formation to track the preset target trajectory, which has extremely high research value.

[0003] Thanks to the high control accuracy and rolling optimization of distributed model predictive control (DMPC), the DMPC strategy is often used for trajectory tracking control of unmanned ship clusters. Compared with centralized MPC, DMPC has multiple sub-controllers and a lower computational burden, thus meeting the real-time requirements of multi-unmanned ship systems. However, the difference between the information sent and received by unmanned ships leads to information gaps within the multi-unmanned ship system, which reduces the control effect of DMPC. In addition, DMPC still requires rolling solution of optimization problems, and its computational complexity cannot meet the needs of some unmanned ships for high-frequency control. Therefore, how to balance the control effect and computational complexity is a major problem in designing a collaborative trajectory tracking control algorithm for multiple unmanned ships.

[0004] The control law learning strategy based on deep learning can use the deep network to output the optimal control law, avoid solving complex optimization problems, and reduce the calculation time of the control algorithm. Long Short-Term Memory (LSTM) is a time recurrent neural network that is good at solving sequence-to-sequence mapping problems (Sequence to Sequence, Seq2Seq). To this end, the LSTM model is used to learn the mapping relationship between distributed control laws and global control laws, which improves the distributed control effect while reducing the solution time of the optimization algorithm, and realizes the collaborative trajectory tracking control of the unmanned ship cluster. Summary of the invention

[0005] The technical problems to be solved by the present invention are:

[0006] In order to avoid the shortcomings of the prior art, the present invention provides a method for cooperative tracking control of unmanned ship clusters based on model predictive control and control law learning, which is used to solve the problems of high computational complexity and poor control effect of the existing unmanned ship cluster cooperative trajectory tracking control algorithm.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A method for cooperative tracking control of unmanned ship clusters based on model predictive control and control law learning, characterized by comprising:

[0009] Step 1: Establish a trajectory tracking error model for the unmanned ship cluster;

[0010] Step 2: Using the global information of the unmanned ship cluster, a centralized model predictive control algorithm is designed based on the trajectory tracking error model to calculate the optimal global control law;

[0011] Step 3: Using the internal interaction information of the unmanned ship cluster, a dual-mode trajectory tracking control algorithm consisting of a distributed model predictive control algorithm and a feedback control law is designed based on the trajectory tracking error model, and the optimal distributed control law is calculated;

[0012] Step 4: Randomly sample different target trajectories and desired formations, use the distributed control law of the dual-mode trajectory tracking control algorithm as input, and use the global control law calculated by the centralized model predictive control algorithm as output to construct a training set for the sequence-to-sequence problem; calculate the tracking eigenvalues ​​based on the target trajectory and the desired formation, and classify the training set based on the tracking eigenvalues;

[0013] Step 5: Establish a control law learning strategy based on the LSTM model to learn the deep mapping relationship between the global control law and the distributed control law; train the LSTM models corresponding to the tracking feature values ​​of different categories based on the classified training set;

[0014] Step 6: After receiving the target trajectory and expected formation, select the optimal LSTM model according to the tracking feature value; calculate the approximate global control law in real time based on the distributed control law, and control the unmanned ship cluster to track the time-varying trajectory based on the approximate global control quantity.

[0015] A further technical solution of the present invention: In step 1, a trajectory tracking error model of a multi-unmanned ship system is established, specifically:

[0016] First, define the coordinates of the unmanned ship i in the world coordinate system as [x s,i ,y s,i ] T , the heading angle is θ s,i , then, the kinematic and dynamic models of the unmanned ship are established

[0017]

[0018] Among them, q s =[x s ,y s ,θ s ] T is the state of an unmanned ship, v x,s ,v y,s ,w s is the linear velocity and angular velocity of the unmanned ship. Assuming that the models of all unmanned ships in the multi-unmanned ship system are the same, the speed update equation of unmanned ship i is:

[0019]

[0020] Among them, m1, m2, m3, d1, d2, d3 are the internal parameters of the unmanned ship related to the additional mass and damping, F x,i is the forward thrust, F w,i is the yaw moment;

[0021] Define the bow state of unmanned ship i as q h,i =[x h,i ,y h,i ,θ i ] T , then the bow kinematic model is

[0022]

[0023] Among them, h is the length from the midpoint of the unmanned ship to the bow; when the unmanned ship i tracks the target trajectory, the kinematic equation of the virtual tracking target is as follows:

[0024]

[0025] in, is the linear velocity and angular velocity of the virtual tracking target; the tracking error is defined as Its update equation is

[0026]

[0027] in, The angle difference between the unmanned ship i and the virtual tracking target is θ e,i =θ r -θ i ; Define the discretized tracking error as:

[0028] q e,i (k+1)=f e (q e,i (k),u e,i (k))δ+q e,i (k)

[0029] =A(w i (k))q e,i (k)+Bu e,i (k)

[0030] in, B = δ, δ is the control interval,

[0031] A further technical solution of the present invention: In step 3, a distributed model predictive control algorithm is designed, specifically:

[0032] The optimization problem of distributed model predictive control is as follows:

[0033]

[0034] st:q h,i (k|k)=η h,i (k),q e,i (k+n|k)=f e (q h,i (k+n|k),q r (k+n))

[0035]

[0036] Among them, q h,i (k+n|k) and q e,i (k+n|k) are respectively based on the control input u i (k+n|k) and linear velocity v y,i (k+n|k) predicted unmanned ship bow state and unmanned ship tracking error; and According to the feasible solution and line speed The predicted bow coordinates and tracking error of unmanned ship j, Q, R, P are weight matrices.

[0037] A further technical solution of the present invention: In step 3, a dual-mode trajectory tracking control algorithm consisting of a distributed model predictive control algorithm and a feedback control law is designed, specifically:

[0038] In a distributed framework, the optimal control input is calculated using k-1 time And the feedback control law constructs a feasible solution as shown below:

[0039]

[0040] in, is the tracking control law, which is defined as:

[0041]

[0042] When the unmanned ship tracking error enters the terminal invariant set, the feedback control law is: The designs of feedback gain K and terminal invariant set Ω are as follows: i and η e ∈Ω, inequality (A K (w i )) T PA K (w i )-P≤-Q * Heng established, where Q * =Q+K T RK, A K (w i )=A(w i )+BK.

[0043] A further technical solution of the present invention: In step 4, the tracking feature value is calculated based on the target trajectory and the expected formation, specifically:

[0044]

[0045] Among them, v r (n) is the speed of the reference trajectory, w r (n) is the angular velocity of the tracking target, d x,i (n), d y,i (n) is the expected distance between the unmanned ship and the desired tracking target.

[0046] A further technical solution of the present invention: In step 4, the training set is classified based on the tracking feature value as follows:

[0047] Randomly select M eigenvalues ​​from the training set as cluster centers, and divide the training set into M categories based on the Euclidean distance between the remaining eigenvalues ​​and the cluster centers;

[0048] Recalculate the center point of each cluster Where N D,j is the dataset D of the jth category j The number of j is the center point of category j;

[0049] Continue classification based on the new center point until the cluster center no longer changes significantly.

[0050] A further technical solution of the present invention: In step 5, a control law learning strategy based on the LSTM model is established to learn the deep mapping relationship between the global control law and the distributed control law, specifically:

[0051] In the training set, the global control law is defined as The distributed control law is is the tracking error predicted by the distributed control law, is the tracking error of unmanned ship j predicted based on the cluster information received by unmanned ship i and the distributed control law;

[0052] The mapping relationship between tracking error and distributed control law and global control law is learned through the encoder-decoder model based on LSTM network:

[0053]

[0054] The output of the LSTM model is the deviation between the global control law and the distributed control law. By minimizing the mean square error (MSE) loss function, the mapping relationship between the distributed control law and the global control law is found.

[0055] A further technical solution of the present invention is as follows: in step 6, after receiving the target trajectory and the desired formation, the optimal LSTM model is selected according to the tracking characteristic value; the approximate global control law is calculated in real time based on the distributed control law, and the unmanned ship cluster is controlled to track the time-varying trajectory based on the approximate global control amount, specifically:

[0056] The unmanned ship i extracts a sub-target trajectory of length N, based on the tracking eigenvalue β of the trajectory L,i Select the optimal LSTM model;

[0057] Using dual-mode trajectory tracking control algorithm, real-time output of distributed control law and predict the states and control inputs of other unmanned vessels;

[0058] Using LSTM model to estimate the difference between global control law and distributed control law Thus, the approximate global control law is calculated in real time

[0059] Finally, according to the first control input in the approximate global control law Calculate the forward thrust F of the unmanned ship x,i and yaw moment F w,i , realizing the trajectory tracking control of unmanned ship i, thereby completing the collaborative trajectory tracking control of the unmanned ship cluster.

[0060] A computer system, characterized in that it comprises: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0061] A computer-readable storage medium, characterized in that it stores computer-executable instructions, which are used to implement the above method when executed.

[0062] The beneficial effects of the present invention are:

[0063] The present invention provides a method for cooperative tracking control of unmanned ship clusters based on model predictive control and control law learning. First, a feedback control law is designed based on the DMPC algorithm, a dual-mode trajectory tracking control framework is constructed, and an LSTM model is designed to learn the deep mapping relationship between the global control law calculated by the centralized MPC and the distributed control law, and finally a control law learning strategy based on the LSTM model and model predictive control is formed. The present invention uses the LSTM model and feedback control law to obtain an approximate global control law without directly calculating the rolling optimization problem, which solves the problem of high computational complexity of the traditional MPC algorithm and improves the control effect of the distributed control algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0065] Figure 1 is a flow chart of an embodiment of the present invention.

[0066] Figure 2 Schematic diagram of collaborative trajectory tracking of unmanned ship clusters.

[0067] Figure 3 Schematic diagram of the collaborative tracking of time-varying trajectories by a swarm of unmanned vessels.

[0068] Figure 4 Schematic diagram of a swarm of unmanned vessels collaboratively tracking a time-varying trajectory in a time-varying formation. DETAILED DESCRIPTION

[0069] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0070] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0071] In order to enable those skilled in the art to better understand the present invention, the present invention is described in detail below in conjunction with specific embodiments.

[0072] Embodiment 1:

[0073] The unmanned ship swarm cooperates to track a preset target track in a fixed formation, including the following steps:

[0074] Step 1: Establish the trajectory tracking error model of the multi-unmanned ship system:

[0075] First, define the coordinates of the unmanned ship i in the world coordinate system as [x s,i ,y s,i ] T , the heading angle is θ s,i , then, the kinematic and dynamic models of the unmanned ship are established

[0076]

[0077] Among them, q s,i =[x s,i ,y s,i ,θ s,i ] T is the state of an unmanned ship, v x,i ,v y,i ,w i is the linear velocity and angular velocity of the unmanned ship. Assuming that the models of all unmanned ships in the multi-unmanned ship system are the same, the speed update equation of unmanned ship i is:

[0078]

[0079] Among them, m1, m2, m3, d1, d2, d3 are the internal parameters of the unmanned ship related to the additional mass and damping, F x,i is the forward thrust, F w,i is the yaw moment.

[0080] Define the bow state of unmanned ship i as q h,i =[x h,i ,y h,i ,θ i ] T , then the bow kinematic model is

[0081]

[0082] Among them, h is the distance from the middle point of the unmanned ship to the bow, and the control input u i =[v x,i ,w i ] T When the unmanned ship i tracks the target trajectory, the kinematic equation of the virtual tracking target is as follows:

[0083]

[0084] in, is the linear velocity and angular velocity of the virtual tracking target. The tracking error is defined as Its update equation is

[0085]

[0086] in, The angle difference between the unmanned ship i and the virtual tracking target is θ e,i =θ r -θ i In order to configure the algorithm in an embedded system, the discretized tracking error is defined as

[0087]

[0088] in, B = δ, δ is the control,

[0089] Step 2: Using the global information of the multi-unmanned ship system, design an unmanned ship cluster collaborative trajectory tracking algorithm based on centralized model predictive control, and calculate the optimal cluster collaborative control law;

[0090] In the offline stage, a centralized cost function is constructed based on the global information within the multi-unmanned ship system:

[0091]

[0092] in, is the number of unmanned ships in the unmanned ship cluster, is the collection of unmanned ships except unmanned ship i, and According to the control input and line speed Predicted unmanned ship bow state and unmanned ship tracking error. At the same time, the tracking errors of all unmanned ships are also updated. and are the tracking error and control input constraints of the unmanned ship, Q, R, Q c and P are weight matrices, and N is the prediction horizon of model predictive control. It is a terminal invariant set, indicating that the tracking errors of all unmanned ships will enter the terminal invariant set at the terminal of the prediction time domain.

[0093] Step 3: Using the interaction information between ships, a dual-mode trajectory tracking control framework consisting of a distributed model predictive control algorithm and a feedback control law is designed;

[0094] During online calculation, due to the influence of the internal information transmission time of the multi-unmanned ship system, the unmanned ship cannot obtain the status of other unmanned ships in the cluster in real time. In order to make full use of the computing power of each unmanned ship, the optimization problem of distributed model predictive control is as follows:

[0095]

[0096] Among them, q h,i (k+n|k) and q e,i (k+n|k) are respectively based on the control input u i (k+n|k) and linear velocity v y,i (k+n|k) predicted unmanned ship bow state and unmanned ship tracking error. and According to the feasible solution and line speed Predicted bow coordinates and tracking error of unmanned ship j. Different from centralized MPC, the states of other unmanned ships in distributed MPC are predicted based on the cluster information received by unmanned ship i at the current moment.

[0097] Step 4: Randomly sample different trajectories and desired formations, use the distributed control law calculated by the dual-mode trajectory tracking control algorithm as input, and the global control law calculated by the centralized model predictive control as output, to construct a training set for the sequence-to-sequence problem (Seq2Seq).

[0098] The distributed control law calculated by randomly sampling the target trajectory using the dual-mode trajectory tracking control framework is defined as is the tracking error predicted by the distributed control law, is the tracking error of unmanned ship j predicted by the cluster information received by unmanned ship i and the distributed control law. For the same trajectory, the optimal global control law calculated using centralized MPC and cluster information is Afterwards, the distributed control law Its predicted tracking error Tracking error with other unmanned vessels As input, the global control law Construct a training set for the output.

[0099] Step 5: Establish a control law learning strategy based on the LSTM model, learn the deep mapping relationship between the global control law and the distributed control law, and design an adaptive model switching strategy for the time-varying target trajectory;

[0100] Step 5.1: Design an encoder-decoder model structure based on the LSTM network to solve the Seq2Seq problem in step 4;

[0101] In the distributed framework, the optimal control input u calculated at the last moment is used i * (k+n|k-1) and the feedback control law construct the feasible solution shown below.

[0102]

[0103] in, For tracking control law, it is defined as:

[0104]

[0105] When the unmanned ship tracking error enters the terminal invariant set, the feedback control law is: The designs of feedback gain K and terminal invariant set Ω are as follows: i and η e ∈Ω, inequality (A K (w i )) T PA K (w i )-P≤-Q * Heng established, where Q * =Q+K T RK, A K (w i )=A(w i )+BK.

[0106] Afterwards, the mapping relationship between tracking error and distributed control law and global control law is learned through the encoder-decoder model based on LSTM network.

[0107]

[0108] The output of the LSTM model is the deviation between the global control law and the distributed control law. By minimizing the mean square error (MSE) loss function, the mapping relationship between the distributed control law and the global control law is found.

[0109] Step 5.2: Analyze the unmanned ship model in step 1 and design tracking feature values ​​related to the target trajectory and the desired formation;

[0110] Aiming at the problem of tracking time-varying target trajectories, an adaptive model switching mechanism is designed to extract a sub-target trajectory η 1→N,r (k), whose length is equal to the prediction time domain of the model predictive control. Then, the most appropriate control strategy is selected according to the maximum speed and expected formation of the sub-target trajectory, and the tracking characteristic value is defined as

[0111]

[0112] Among them, v r (n) is the speed of the reference trajectory, w r (n) is the angular velocity of the tracking target, d x,i (n), d y,i (n) are the x and y distances between the unmanned ship and the desired tracking target.

[0113] Step 5.3: Use the clustering algorithm to classify the data set in step 4 according to the tracking feature value, then train different categories of LSTM models respectively, and select the appropriate LSTM model according to the tracking feature value;

[0114] First, randomly select M cluster centers from the data and divide the training set into M categories based on the Euclidean distance. Then, recalculate the center point of each cluster Where N D,j is the dataset D of the jth category j The number of j is the center point of category j. Finally, continue to divide and classify according to the new center point, and repeat the above steps until the cluster center no longer changes significantly.

[0115] Step 6: After receiving the target trajectory and the desired formation, the optimal LSTM model is selected according to step 5, and the approximate global control law is calculated in real time based on the distributed control law in step 4 to control the unmanned ship cluster to track the time-varying trajectory;

[0116] First, the unmanned ship i extracts a sub-target trajectory of length N and calculates its tracking eigenvalue β L,i , and select the optimal LSTM model accordingly. Then, using the dual-mode control framework, the distributed control law is output in real time The state and control input of other unmanned ships are predicted based on the received cluster information. Afterwards, the LSTM model is used to estimate the difference between the global control law and the distributed control law. Thus, the approximate global control law is calculated in real time Finally, the first control input in the approximate global control law Calculate the forward thrust F of the unmanned ship x,i and yaw moment F w,i , realizing the trajectory tracking control of unmanned ship i, thereby completing the collaborative trajectory tracking control of the unmanned ship cluster.

[0117] Example 2: Unmanned ship cluster tracking time-varying target trajectory

[0118] like Figure 3 As shown, when controlling the unmanned ship cluster to track a time-varying trajectory, the conventional trajectory tracking algorithm usually relies on the maximum value of the target trajectory speed to design a conservative feedback control law, which reduces the control effect of the trajectory tracking algorithm. To this end, the present invention designs an unmanned ship cluster collaborative tracking control method based on model predictive control and control law learning. During offline training, the training set contains different target trajectories. Then, consistent with the calculation steps in Example 1, the training set is classified by extracting the tracking features of the target trajectory, and then the LSTM model adapted to different tracking target speeds is trained. During online tracking, the tracking target is classified by extracting the features of the tracking trajectory in real time, thereby selecting the optimal LSTM model, and adaptively outputting an approximate global control input, which improves the control effect of the distributed model predictive control algorithm while reducing the optimization complexity and conservatism of the tracking control algorithm.

[0119] Example 3: Unmanned ship clusters tracking time-varying target trajectories in a time-varying formation

[0120] like Figure 4 As shown, when the unmanned ship cluster tracks the time-varying trajectory in a time-varying formation, the conventional trajectory tracking algorithm usually relies on the maximum value of the target trajectory speed and the farthest distance between the unmanned ship and the tracking target in the time-varying formation to design a conservative feedback control law, which reduces the control effect of the trajectory tracking algorithm. To this end, the present invention designs an unmanned ship cluster collaborative tracking control method based on model predictive control and control law learning. During offline training, the training set contains different target trajectories and different collaborative formations. Then, consistent with the calculation steps in Example 1, the training set is classified by extracting the tracking features of the target trajectory and the expected formation, and then the LSTM model adapted to different tracking target speeds and formations is trained.

[0121] During online tracking, the tracking trajectory and the expected formation are classified by extracting the features of the tracking trajectory and the expected formation in real time, so as to select the optimal LSTM model. When designing a dual-mode control framework based on distributed MPC and feedback control law, the feedback control law requires that the tracking error is within the terminal invariant set. When the new expected formation mutation exceeds the terminal invariant set, that is, The feedback control law cannot output a feasible solution. Therefore, the present invention designs a segmented formation transformation strategy, assuming that the expected formation does not change at k-1, and the tracking error is within the terminal invariant set, that is, q e,i (k-1)∈Ω, then at time k, the formation transformation needs to ensure that the tracking error is within the terminal invariant set, and the tracking target state is defined as η r (k) = f r (η r (k-1),u r (k-1))+Δd, where the formation transformation distance Δd must satisfy Δd∈{d e |f e (q h,i (k),q r (k)+d e )∈Ω}. After that, the target trajectory and the tracking features of the expected formation are extracted in real time, and the optimal LSTM model is selected based on the trajectory features. Finally, the approximate global control input is calculated in accordance with the real-time example 1, thereby controlling the unmanned ship cluster to track the time-varying trajectory in a time-varying formation.

[0122] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should be included in the protection scope of the present invention.

Claims

1. A method for cooperative tracking control of unmanned ship clusters based on model predictive control and control law learning, characterized in that: include: Step 1: Establish a trajectory tracking error model for the unmanned ship cluster; Step 2: Using the global information of the unmanned ship cluster, a centralized model predictive control algorithm is designed based on the trajectory tracking error model to calculate the optimal global control law; Step 3: Using the internal interaction information of the unmanned ship cluster, a dual-mode trajectory tracking control algorithm consisting of a distributed model predictive control algorithm and a feedback control law is designed based on the trajectory tracking error model, and the optimal distributed control law is calculated; Step 4: Randomly sample different target trajectories and desired formations, use the distributed control law of the dual-mode trajectory tracking control algorithm as input, and use the global control law calculated by the centralized model predictive control algorithm as output to construct a training set for the sequence-to-sequence problem; calculate the tracking eigenvalues ​​based on the target trajectory and the desired formation, and classify the training set based on the tracking eigenvalues; Step 5: Establish a control law learning strategy based on the LSTM model to learn the deep mapping relationship between the global control law and the distributed control law; train the LSTM models corresponding to the tracking feature values ​​of different categories based on the classified training set; Step 6: After receiving the target trajectory and expected formation, select the optimal LSTM model according to the tracking feature value; calculate the approximate global control law in real time based on the distributed control law, and control the unmanned ship cluster to track the time-varying trajectory based on the approximate global control quantity.

2. According to claim 1, a method for cooperative tracking control of unmanned ship clusters based on model predictive control and control law learning is characterized in that: In step 1, a trajectory tracking error model of a multi-unmanned ship system is established, specifically: First, define the coordinates of the unmanned ship i in the world coordinate system as [x s,i ,y s,i ] T , the heading angle is θ s,i , then, the kinematic and dynamic models of the unmanned ship are established Among them, q s =[x s ,y s ,θ s ] T is the state of an unmanned ship, v x,s ,v y,s ,w s is the linear velocity and angular velocity of the unmanned ship. Assuming that the models of all unmanned ships in the multi-unmanned ship system are the same, the speed update equation of unmanned ship i is: Among them, m1, m2, m3, d1, d2, d3 are the internal parameters of the unmanned ship related to the additional mass and damping, F x,i is the forward thrust, F w,i is the yaw moment; Define the bow state of unmanned ship i as q h,i =[x h,i ,y h,i ,θ i ] T , then the bow kinematic model is Among them, h is the length from the midpoint of the unmanned ship to the bow; when the unmanned ship i tracks the target trajectory, the kinematic equation of the virtual tracking target is as follows: in, is the linear velocity and angular velocity of the virtual tracking target; the tracking error is defined as Its update equation is in, The angle difference between the unmanned ship i and the virtual tracking target is θ e,i =θ r -θ i ; Define the discretized tracking error as: q e,i (k+1)=f e (q e,i (k),u e,i (k))δ+q e,i (k) <h2 style=";text-align:left;direction:ltr">=A(w<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (k))q<h2 style=";text-align:left;direction:ltr"> e,i <h2 style=";text-align:left;direction:ltr"> (k)+Bu<h2 style=";text-align:left;direction:ltr"> e,i <h2 style=";text-align:left;direction:ltr"> (k) in, B = δ, δ is the control interval, 3. According to claim 2, a method for cooperative tracking control of unmanned ship clusters based on model predictive control and control law learning is characterized in that: In step 3, a distributed model predictive control algorithm is designed, specifically: The optimization problem of distributed model predictive control is as follows: st:q h,i (k|k)=η h,i (k),q e,i (k+n|k)=f e (q h,i (k+n|k),q r (k+n)) q h,i (k+n+1|k)=f h (q h,i (k+n|k),u i (k+n|k),v y,i (k+n|k)) Among them, q h,i (k+n|k) and q e,i (k+n|k) are respectively based on the control input u i (k+n|k) and linear velocity v y,i (k+n|k) predicted unmanned ship bow state and unmanned ship tracking error; and According to the feasible solution and line speed The predicted bow coordinates and tracking error of unmanned ship j, Q, R, P are weight matrices.

4. According to claim 3, a method for cooperative tracking control of unmanned ship clusters based on model predictive control and control law learning is characterized in that: In step 3, a dual-mode trajectory tracking control algorithm consisting of a distributed model predictive control algorithm and a feedback control law is designed, specifically: In a distributed framework, the optimal control input is calculated using k-1 time And the feedback control law constructs a feasible solution as shown below: in, is the tracking control law, which is defined as: When the unmanned ship tracking error enters the terminal invariant set, the feedback control law is: The designs of feedback gain K and terminal invariant set Ω are as follows: i and η e ∈Ω, inequality (A K (w i )) T PA K (w i )-P≤-Q * Heng established, where Q * =Q+K T RK, A K (w i )=A(w i )+BK.

5. The unmanned ship cluster collaborative tracking control method based on model predictive control and control law learning according to claim 4 is characterized in that: In step 4, the tracking feature value is calculated based on the target trajectory and the expected formation, specifically: Among them, v r (n) is the speed of the reference trajectory, w r (n) is the angular velocity of the tracking target, d x,i (n), d y,i (n) is the expected distance between the unmanned ship and the desired tracking target.

6. According to claim 5, a method for cooperative tracking control of unmanned ship clusters based on model predictive control and control law learning is characterized in that: In step 4, the training set is classified based on the tracking feature value as follows: Randomly select M eigenvalues ​​from the training set as cluster centers, and divide the training set into M categories based on the Euclidean distance between the remaining eigenvalues ​​and the cluster centers; Recalculate the center point of each cluster Where N D,j is the dataset D of the jth category j The number of j is the center point of category j; Continue classification based on the new center point until the cluster center no longer changes significantly.

7. The unmanned ship cluster cooperative tracking control method based on model predictive control and control law learning according to claim 6 is characterized in that: In step 5, a control law learning strategy based on the LSTM model is established to learn the deep mapping relationship between the global control law and the distributed control law, specifically: In the training set, the global control law is defined as The distributed control law is is the tracking error predicted by the distributed control law, is the tracking error of unmanned ship j predicted based on the cluster information received by unmanned ship i and the distributed control law; The mapping relationship between tracking error and distributed control law and global control law is learned through the encoder-decoder model based on LSTM network: The output of the LSTM model is the deviation between the global control law and the distributed control law. By minimizing the mean square error (MSE) loss function, the mapping relationship between the distributed control law and the global control law is found.

8. The unmanned ship cluster cooperative tracking control method based on model predictive control and control law learning according to claim 7 is characterized in that: In step 6, after receiving the target trajectory and the desired formation, the optimal LSTM model is selected according to the tracking feature value; the approximate global control law is calculated in real time based on the distributed control law, and the unmanned ship cluster is controlled to track the time-varying trajectory based on the approximate global control amount, specifically: The unmanned ship i extracts a sub-target trajectory of length N, based on the tracking eigenvalue β of the trajectory L,i Select the optimal LSTM model; Using dual-mode trajectory tracking control algorithm, real-time output of distributed control law and predict the states and control inputs of other unmanned vessels; Using LSTM model to estimate the difference between global control law and distributed control law Thus, the approximate global control law is calculated in real time Finally, according to the first control input in the approximate global control law Calculate the forward thrust F of the unmanned ship x,i and yaw moment F w,i , realizing the trajectory tracking control of unmanned ship i, thereby completing the collaborative trajectory tracking control of the unmanned ship cluster.

9. A computer system, characterized in that include: One or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method of claim 1.

10. A computer-readable storage medium, characterized in that Computer executable instructions are stored, and when the instructions are executed, they are used to implement the method of claim 1.

Images