Water surface unmanned ship multi-mode prediction control method based on deep learning assistance

Abstract

The invention discloses a water surface unmanned ship multi-mode prediction control method based on deep learning assistance, and the method comprises the steps: building a corresponding unmanned ship nonlinear motion model, and enabling an unmanned ship to comprise a plurality of operation modes and sub-modes facing different task types; according to an actual application scene, constructing a preset nonlinear model prediction controller; acquiring a training data set through analog simulation considering environment disturbance and obstacle constraint; training the LSTM network to obtain a learning model; in the operation stage, the prediction state of the unmanned ship is obtained through the learning model, and state deviation is quantified; and inputting the state deviation into a preset nonlinear model prediction controller, obtaining corrected single-step optimal control input by solving a single-step optimal control problem, and driving the unmanned ship. Based on the deep learning method, the influence of environment disturbance and mode change can be reduced, and safe and reliable control performance is provided in different modes and complex environments.

Description

Technical Field

[0001] The present invention belongs to the technical field of automatic control of unmanned intelligent ships, and in particular relates to a multi-mode predictive control method for an unmanned surface ship based on deep learning assistance. Background Art

[0002] Unmanned surface vehicles (USVs) have been widely used in various fields such as security patrols, environmental monitoring, resource exploration, and maritime rescue. Compared with traditional ships, USVs have obvious advantages such as low cost, high efficiency, and low risk. At present, research in the field of unmanned ships has become a research hotspot in the field of shipbuilding and marine engineering.

[0003] The control technology in motion stability and tracking is an important issue in the motion control of unmanned ships. In narrow waterways or complex waters, there are many ships and obstacles, the traffic density is high, and the degree of danger is high. Safety collision avoidance (CA) should also be considered. For local collision avoidance problems, integrating the CA scheme into the controller can ensure the safety of the ship itself and other obstacles. Model Predictive Control (MPC) is an efficient strategy for controlling nonlinear multi-input and multi-output systems, which is mainly divided into linear and nonlinear. For systems with highly nonlinear, time-varying and complex dynamic characteristics, nonlinear models are a more appropriate choice, and the safety control and collision avoidance of unmanned ships often involve many constraints.

[0004] Multi-mode control is a control strategy that changes with the operating state of the system. It can achieve the combined advantages of multiple control strategies and optimize the control performance of complex systems. At present, some improved MPC methods designed to handle multi-mode processes are called multi-mode MPC or switching MPC. These methods establish multiple models for each operating mode and design model switching strategies. Some people have achieved good control effects in four-wheel steering vehicles based on multi-mode control strategies of fuzzy selectors. Some people have designed adaptive multi-mode power control algorithms for microgrid direct-drive permanent magnet wind power generation systems. Some people have proposed a nonlinear pH neutralization process MPC based on multi-model neural networks. Some people have designed a soft switching scheme for autonomous underwater vehicles to obtain better control performance. Due to the existence of complex coupling nonlinearities and large time lags, the motion characteristics of USVs will fluctuate greatly under different mission operating conditions. Therefore, it is necessary to study the multi-mode control strategy of USVs to ensure their effective operational adaptability.

[0005] There is little research on multi-mode control in USV applications. Although significant progress has been made in various aspects of developing unmanned ships, it is important to fully realize the potential of multi-mode control strategies and design appropriate algorithms for predicting the behavior of the controlled system to automatically match different operating modes without switching strategies. Summary of the invention

[0006] The purpose of the present invention is to provide a multi-mode predictive control method for an unmanned surface ship based on deep learning assistance to address the deficiencies of the prior art. The present invention is based on a deep learning method, can reduce the impact of environmental disturbances and mode changes, and provide safe and reliable control performance in different modes and complex environments.

[0007] The objective of the present invention is achieved through the following technical solutions: In a first aspect, an embodiment of the present invention provides a multi-mode predictive control method for an unmanned surface ship based on deep learning assistance, comprising:

[0008] Establishing a nonlinear motion model of a surface unmanned ship; wherein the unmanned ship includes multiple operation modes for different mission types, and each operation mode has multiple sub-modes;

[0009] Constructing a preset nonlinear model predictive controller for the unmanned surface ship;

[0010] Add obstacle avoidance constraints, and record the system status and control input data of the surface unmanned ship in real time during the simulation process with environmental disturbances or measurement errors to obtain a training data set;

[0011] Using the training data set to train the long short-term memory network through a preset deep learning training algorithm, and obtaining a learning model after the training is completed;

[0012] Based on the learning model and the nonlinear model predictive controller, the learning model is used in the operation phase to obtain the predicted state of the surface unmanned ship and quantify the state deviation;

[0013] The state deviation is input into the preset nonlinear model predictive controller, and the corrected single-step optimal control input is obtained by solving the single-step optimal control problem, and the surface unmanned vessel is driven;

[0014] Repeat the above process and perform multi-mode predictive control of the surface unmanned vessel through a deep learning-assisted control model.

[0015] Furthermore, the nonlinear motion model of the surface unmanned ship is established; wherein the unmanned ship includes multiple operation modes for different task types, and each operation mode has multiple sub-modes, specifically including:

[0016] A fixed frame coordinate system tangent to the earth's surface and a motion coordinate system fixed on the hull are established to represent the kinematic model and dynamic model of the surface unmanned ship, which is the nonlinear motion model of the surface unmanned ship.

[0017] According to different mission types, the control mode of the unmanned ship is divided into four operating modes: straight sailing, turning, point stabilization and trajectory tracking. On the basis of the above four operating modes, each operating mode is subdivided into two modes: speed and collision. Among them, the speed mode includes three types: low speed, medium speed and high speed. The collision mode includes four types: static obstacle, overtaking, encounter and cross encounter according to the obstacle type and the encounter situation of the ship.

[0018] Furthermore, the nonlinear motion model of the unmanned surface ship includes a kinematic model and a dynamic model, and the expressions thereof are respectively:

[0019]

[0020] Where M1, M2, M3 and D1, D2, D3 are the diagonal elements of the inertia matrix and the diagonal linear components of the damping matrix respectively; x, y, ψ represent the x-axis position, y-axis position and direction in the fixed coordinate system respectively; Respectively represent the instantaneous changes in the x-axis position, y-axis position and direction in the fixed coordinate system; u, v, r represent the x-axis velocity, y-axis velocity and angular velocity in the moving coordinate system; Respectively represent the instantaneous changes of x-axis velocity, y-axis velocity and angular velocity in the motion coordinate system; τ u and τ r They represent the force and torque provided by the two propellers respectively.

[0021] Furthermore, the construction of a preset nonlinear model predictive controller for the unmanned surface ship specifically includes:

[0022] Based on the nonlinear system, a nonlinear model predictive controller is constructed for the surface unmanned ship. The nonlinear model predictive controller solves the optimal control problem in the future finite prediction time domain at each moment at the cost of the deviation between the system state and the reference trajectory and the control input increment:

[0023]

[0024] Among them, J is the cost function, is the deviation between the system state and the reference trajectory, x k is the system state variable, is the reference target or trajectory, k represents the kth moment, k+i represents the k+ith moment, f(·) represents the nonlinear system, x min ,x maxis the constraint condition of the system state variables, u min ,u max is the system control variable constraint, H represents the finite prediction horizon, represents the quadratic norm x T Qx, represents the quadratic norm x T Rx, Q≥0, R≥0 is an adjustable weight matrix.

[0025] Furthermore, the obstacle avoidance constraint is added by recording the system state and control input data of the surface unmanned ship in real time during a simulation process with environmental disturbances or measurement errors to obtain a training data set, specifically including:

[0026] Based on the preset nonlinear model predictive controller constructed for the surface unmanned vessel, the corresponding obstacle avoidance constraints are added to the nonlinear model predictive control;

[0027] In the simulation process, environmental disturbances or measurement errors are added to obtain and record the system state and control input data of the surface unmanned ship in real time to obtain a first data set;

[0028] The first data set is preprocessed to obtain a training data set; wherein the preprocessing includes importing, extracting features and labels, dividing the data set, normalizing and packaging the data.

[0029] Furthermore, based on the learning model and the nonlinear model predictive controller, the learning model is used in the operation phase to obtain the predicted state of the surface unmanned ship and quantify the state deviation, which specifically includes:

[0030] Based on the trained learning model, the predicted state of the surface unmanned vessel is obtained according to the historical system state, historical control input data and the current system state and current control input data;

[0031] According to the predicted state and actual state of the surface unmanned ship, the state deviation of the surface unmanned ship, that is, the state correction data, is calculated.

[0032] Furthermore, the state deviation is input into a preset nonlinear model predictive controller, and a corrected single-step optimal control input is obtained by solving a single-step optimal control problem, and the unmanned ship is driven, specifically including:

[0033] The state correction data is input into the preset nonlinear model predictive controller, and the single-step optimal control problem is solved by adjusting the system state and control input weight of the surface unmanned ship, thereby calculating the optimal control input data at the current moment;

[0034] Applying the optimal control input data to the unmanned surface ship to drive the unmanned surface ship to move according to the revised control strategy;

[0035] In each control cycle, the corrected control input data and the real-time system status of the surface unmanned vessel are recorded.

[0036] A second aspect of an embodiment of the present invention provides a multi-mode predictive control system for an unmanned surface ship assisted by deep learning, which is used to implement the multi-mode predictive control method for an unmanned surface ship assisted by deep learning, including:

[0037] Establishing a module for establishing a nonlinear motion model of a surface unmanned ship; wherein the unmanned ship includes multiple operation modes for different task types, and each operation mode has multiple sub-modes;

[0038] A first construction module is used to construct a preset nonlinear model predictive controller for the unmanned surface ship;

[0039] The sampling combination module is used to add obstacle avoidance constraints and record the system status and control input data of the surface unmanned ship in real time during the simulation process with environmental disturbances or measurement errors to obtain the training data set;

[0040] A second construction module is used to train the long short-term memory network using the training data set through a preset deep learning training algorithm, and obtain a learning model after the training is completed;

[0041] An update module is introduced, which is used to obtain the predicted state of the surface unmanned ship by using the learning model and the nonlinear model predictive controller during the operation phase, and quantify the state deviation;

[0042] A calculation module, used for inputting the state deviation into a preset nonlinear model predictive controller, obtaining a corrected single-step optimal control input by solving the single-step optimal control problem, and driving the surface unmanned vessel; and

[0043] An implementation module is used to repeatedly call the above modules and perform multi-mode predictive control of surface unmanned ships through a deep learning-assisted control model.

[0044] A third aspect of an embodiment of the present invention provides an electronic device, including:

[0045] one or more processors; and

[0046] a memory coupled to the processor;

[0047] Among them, the memory is used to store program data, and the processor is used to execute the program data to implement the above-mentioned multi-mode predictive control method of the surface unmanned ship assisted by deep learning.

[0048] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, are used to implement the above-mentioned multi-mode predictive control method for an unmanned surface ship assisted by deep learning.

[0049] The beneficial effects of the present invention are as follows: first, the present invention establishes a mathematical model of nonlinear motion of a surface unmanned ship, wherein the unmanned ship includes multiple operating modes for different task types, and each operating mode has several seed modes; then, a preset nonlinear model predictive controller is constructed according to the actual application scenario to adapt to the navigation of the unmanned ship in different modes in complex and changing environments, and in combination with the corresponding obstacle avoidance constraints, a training data set such as the state and input of the unmanned ship is obtained in real time through a simulation process in which there is environmental disturbance or measurement error; then, the LSTM network is trained according to the training data set through a preset deep learning training algorithm to obtain a learning model; further, in the actual operation stage, the learning model is used to obtain the predicted state of the unmanned ship, and the state deviation is quantified, and the state deviation is input into the preset nonlinear model predictive controller, and the single-step optimal control problem is solved to obtain a corrected single-step optimal control input, and the unmanned ship is driven based on the corrected control input data; the present invention uses a deep learning-assisted control model to perform multi-mode predictive control of a surface unmanned ship, which can provide safe and reliable control performance in different modes and complex environments, and is conducive to reducing the impact of environmental disturbances and mode changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of a multi-mode predictive control method for an unmanned surface ship based on deep learning assistance of the present invention;

[0051] Figure 2 The present invention Figure 1 Implementation flow chart of step S1 in ;

[0052] Figure 3 The present invention Figure 1 Implementation flow chart of step S5 in ;

[0053] Figure 4 The present invention Figure 1 Implementation flow chart of step S7 in FIG.

[0054] Figure 5 It is a structural block diagram of a multi-mode predictive control system of an unmanned surface ship based on deep learning assistance of the present invention;

[0055] Figure 6 It is a structural schematic diagram of the electronic device of the present invention. DETAILED DESCRIPTION

[0056] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0057] The terms used in the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "the" and "the" used in the present invention and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0058] It should be understood that although the terms first, second, third, etc. may be used in the present invention to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present invention, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0059] The present invention is described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the features of the following embodiments and implementations can be combined with each other.

[0060] See also Figure 1 The multi-mode predictive control method of an unmanned surface ship based on deep learning assistance of the present invention specifically comprises the following steps:

[0061] S1. Establish a nonlinear motion model of a surface unmanned ship. The unmanned ship includes multiple operation modes for different mission types, and each operation mode has multiple sub-modes.

[0062] like Figure 2 As shown, it specifically includes the following sub-steps:

[0063] S11. Establish a fixed frame coordinate system tangent to the earth's surface and a motion coordinate system fixed on the hull to represent the kinematic model and dynamic model of the surface unmanned ship, that is, the nonlinear motion model of the surface unmanned ship.

[0064] Specifically, a nonlinear motion model of a surface unmanned ship is established, wherein the surface unmanned ship is usually underdriven, that is, driven by one propeller and one rudder, or driven by two propellers. In this embodiment, the following description is based on the drive by two propellers. The force and torque provided by the two propellers are τ u and τ r The force of the left and right propellers (F l and F r ) can be obtained by formula F l =(τ u +τ r / d m ) / 2 and F r =(τ u -τ r / d m ) / 2, where d m Represents the moment arm of the unmanned ship. The integrated control framework needs to be represented by two coordinate systems, namely the fixed frame coordinate system and the moving coordinate system. The fixed frame coordinate system is tangent to the surface of the earth, and its x-axis and y-axis point to the north and east respectively; the moving coordinate system is connected to the hull, and its axes represent the bow and starboard respectively. Assuming that the unmanned ship has good symmetry, its nonlinear motion model includes a kinematic model and a dynamic model, which are expressed as:

[0065]

[0066] Where M1, M2, M3 and D1, D2, D3 are the diagonal elements of the inertia matrix and the diagonal linear components of the damping matrix respectively; x, y, ψ represent the x-axis position, y-axis position and direction in the fixed coordinate system respectively; Respectively represent the instantaneous changes in the x-axis position, y-axis position and direction in the fixed coordinate system; u, v, r represent the x-axis velocity, y-axis velocity and angular velocity in the moving coordinate system; They represent the instantaneous changes of the x-axis velocity, y-axis velocity and angular velocity in the motion coordinate system respectively. On this basis, the velocity state of the unmanned ship is generated according to the dynamic model (i.e., formula (2)), and then the position state of the unmanned ship is updated by the kinematic model (i.e., formula (1)).

[0067] S12. The control mode of the unmanned ship is divided into four operating modes: straight sailing, turning, point stabilization and trajectory tracking according to different mission types. On the basis of the above four operating modes, each operating mode is further divided into two modes: speed and collision. Among them, the speed mode includes three types: low speed, medium speed and high speed. The collision mode includes four types: static obstacle, overtaking, encounter and cross encounter according to the obstacle type and the encounter situation of the ship.

[0068] Specifically, the control mode is set to It can be expressed as:

[0069]

[0070] in, There are four operating modes: straight flight, turning, point stabilization and trajectory tracking; Indicates the sub-mode under each operating mode, s=1,2,3 indicates three speed types: low speed, medium speed and high speed, c=1,2,…,4 indicates four types: static obstacle, overtaking, encounter and cross encounter according to the obstacle type and ship encounter situation.

[0071] S2. Construct a preset nonlinear model predictive controller for the surface unmanned ship.

[0072] Specifically, according to the actual application scenarios, model predictive control is mainly divided into linear and nonlinear. For systems with highly nonlinear, time-varying and complex dynamic characteristics, nonlinear models are a more appropriate choice. Surface unmanned ships have highly nonlinear, time-varying and complex dynamic characteristics, and surface unmanned ship motion control and collision avoidance problems often involve many constraints. Therefore, consider the following nonlinear system:

[0073]

[0074] Among them, x k is the system state variable at time k, u k is the system control variable at time k, and f(·) represents a nonlinear system. Therefore, according to the actual application scenario, a nonlinear model predictive controller is constructed for the surface unmanned ship based on the above nonlinear system. The nonlinear model predictive controller solves the optimal control problem in the future finite prediction time domain at each moment at the cost of the deviation between the system state and the reference trajectory and the control input increment:

[0075]

[0076] Among them, J is the cost function, is the deviation between the system state and the reference trajectory, x k is the system state variable, is the reference target or trajectory, k represents the kth moment, k+i represents the k+ith moment, f(·) represents the nonlinear system, x min ,x max is the constraint condition of the system state variables, u min ,u max is the system control variable constraint, H represents the finite prediction horizon, and the optimal control sequence in the finite prediction horizon H is expressed as represents the quadratic norm x T Qx, represents the quadratic norm xT Rx, Q≥0, R≥0 is an adjustable weight matrix, and the control requirements for different variables can be changed by adjusting the elements.

[0077] S3. Add obstacle avoidance constraints, and record the system status and control input data of the surface unmanned ship in real time during the simulation process with environmental disturbances or measurement errors to obtain a training data set.

[0078] Specifically, based on the preset nonlinear model predictive controller constructed for the surface unmanned ship, the corresponding obstacle avoidance constraints are added to the nonlinear model predictive control. Assume that the obstacle target (OT) is approximated as a radius of R OT If there is a circular obstacle, the avoidance problem of the controlled ship (Own-ship, OS) can be transformed into:

[0079]

[0080] Among them, R OS D is half the length or breadth of the controlled ship. safe is the safety distance between the obstacle and the outer contour of the ship, R OT is the radius of the obstacle target, which can be set to half the ship length or width according to the shape of the obstacle target; (x(k), y(k)) and (x0(k), y0(k)) are the positions of the controlled ship and the obstacle target in the fixed coordinate system respectively. In the simulation process, environmental disturbances or measurement errors are added, and the system status and control input data of the surface unmanned ship are acquired and recorded in real time to obtain the first data set. The first data set is preprocessed to obtain the training data set; wherein the preprocessing includes importing, extracting features and labels, dividing the data set, normalizing and packaging the data.

[0081] S4. Use the training data set to train the Long Short-Term Memory (LSTM) network through the preset existing deep learning training algorithm, and obtain the learning model after the training is completed.

[0082] Specifically, the LSTM network includes three gating units, followed by a fully connected layer, which simplifies the high-dimensional output into a single dimension to obtain the network output. Since the LSTM network uses a multi-step state prediction system output, the loss function of the LSTM network is designed as:

[0083]

[0084] Among them, L represents the loss function of the LSTM network, is the reference output, is the predicted output. According to the controlled nonlinear system (i.e., formula (4)), the input sequence comes from two parts, namely, the previous system state and the previous control input l k and l u Indicates the lag between system output and input. By incorporating these historical states and control inputs into the LSTM network, the LSTM network is trained and a learning model is obtained after the training is completed. The learning model can capture time dependencies and complex dynamic changes.

[0085] S5. Based on the learning model and the nonlinear model predictive controller, the learning model is used in the operation phase to obtain the predicted state of the surface unmanned vessel and quantify the state deviation.

[0086] like Figure 3 As shown, it specifically includes the following sub-steps:

[0087] S51. Based on the trained learning model, the predicted state of the surface unmanned vessel is obtained according to the historical system state, historical control input data and the current system state and current control input data.

[0088] Specifically, the predicted state of the surface unmanned ship is obtained based on the trained learning model, and its expression is:

[0089]

[0090] in, For the predicted state, is the historical system status, is the historical control input data and g(·) represents the learning model.

[0091] S52. Calculate the state deviation of the unmanned surface ship, that is, the state correction data, according to the predicted state and the actual state of the unmanned surface ship.

[0092] Specifically, the error between the predicted state and the actual state is quantified as follows:

[0093]

[0094] Among them, E k represents the deviation between the predicted state and the actual state of the underwater unmanned ship at the current k moment, For reference status.

[0095] It should be understood that the actual state is the reference state, that is, the state of the reference trajectory. In the task of tracking the trajectory of the surface unmanned ship, the entire trajectory is known at every moment.

[0096] S6. Input the state deviation into the preset nonlinear model predictive controller, obtain the corrected single-step optimal control input by solving the single-step optimal control problem, and drive the surface unmanned vessel.

[0097] Specifically, the state correction data is input into the preset nonlinear model predictive controller, and the single-step optimal control problem is solved by adjusting the system state and control input weight of the surface unmanned ship, thereby calculating the optimal control input data at the current moment. The optimal control input data is applied to the surface unmanned ship to drive the surface unmanned ship to move according to the corrected control strategy:

[0098]

[0099] in, is the optimal control sequence in the finite prediction horizon H The first element in w k+i-1 is the environmental disturbance or measurement error, represents the weight matrix of the n-dimensional system state variables, represents the weight matrix of the unmanned ship’s m-dimensional control input, α n and β m is the weight correction coefficient. In each control cycle, the corrected control input data and the real-time system status of the surface unmanned vessel are recorded for further analysis and evaluation.

[0100] In the above-mentioned surface unmanned vessel system, the mode changes are reflected in the following two aspects: (1) During the operation stage, different system state mode weights can be adjusted in real time according to the speed changes of the unmanned vessel, such as low speed, medium speed, and high speed, to achieve balanced control of speed, position, and angle; (2) By adding obstacle avoidance constraints to the controller, different control input mode weights can be adjusted according to the static or dynamic obstacles of the unmanned vessel during the operation stage to achieve collision avoidance operations that comply with collision avoidance rules, such as right turn avoidance.

[0101] S7. Repeat the process corresponding to the above steps S2 to S6, and perform multi-mode predictive control of the surface unmanned ship through a deep learning-assisted control model.

[0102] In the process of executing the multi-mode predictive control of surface unmanned vessels through deep learning-assisted control models, such as Figure 4 As shown, it specifically includes the following sub-steps:

[0103] S71. Construct a preset nonlinear model predictive controller according to the nonlinear motion model and task type of the surface unmanned vessel.

[0104] Specifically, based on the modeling and system understanding of the unmanned ship, a preset nonlinear model predictive controller is constructed. This controller takes into account the dynamic characteristics and control requirements of the unmanned ship, and can predict and control the motion of the unmanned ship through mathematical modeling and control algorithm design.

[0105] S72. Based on the preset nonlinear model predictive controller constructed for the surface unmanned vessel, corresponding obstacle avoidance constraints are added, and a training data set is obtained through a simulation process with environmental disturbances or measurement errors for training the learning model.

[0106] Specifically, obstacle avoidance constraints are added to the nonlinear model predictive controller to ensure that the unmanned ship can take appropriate avoidance actions. Then, the controller with integrated collision avoidance constraints is used to perform simulations in an environment with disturbances or measurement errors to obtain a training data set. During the simulation, the state and control input of the unmanned ship are recorded to generate a training data set, which is used for training the learning model after preprocessing.

[0107] S73. According to the trained learning model, use the system status and control input data at the historical and current moments to predict the predicted state of the surface unmanned ship at the next moment, calculate the state error of the surface unmanned ship, and obtain the state correction data.

[0108] Specifically, according to the trained learning model, the system state and control input data of the historical and current moments are used to predict the predicted state of the surface unmanned ship at the next moment. The error between the predicted state and the actual state is used to calculate the state correction data.

[0109] S74. Adjust the control input of the unmanned surface ship according to the state correction data, drive the unmanned surface ship to move according to the corrected control strategy, realize multi-mode predictive control of the unmanned surface ship, and record the new control input data and system status in real time for the next step of optimization.

[0110] It can be seen from the above embodiments that the present invention comprehensively utilizes deep learning algorithms and nonlinear model predictive controllers in different operating modes and complex environments, and can adapt to complex and changeable environments and task requirements to improve the accuracy of multi-mode control of surface unmanned ships. Specifically, the present invention establishes a corresponding nonlinear motion model of a surface unmanned ship, and constructs a nonlinear model predictive controller for the unmanned ship to adapt to the navigation of the unmanned ship in complex and changing environments with different operating modes; the present invention obtains a training data set in real time through the established nonlinear system model and simulates the simulation process according to relevant constraints, and trains the LSTM network through a preset deep learning training algorithm according to the training data set to obtain a learning model with accurate prediction performance; the present invention obtains the predicted state of the unmanned ship based on the learning model, and quantifies the state deviation, and inputs the state deviation into the preset nonlinear model predictive controller, and obtains the corrected single-step optimal control input by solving the single-step optimal control problem, and drives the unmanned ship based on the corrected control input data, which effectively solves the problem of comprehensive motion planning and multi-mode control of the surface unmanned ship.

[0111] Corresponding to the aforementioned embodiment of the multi-mode predictive control method for an unmanned surface ship based on deep learning assistance, the present invention also provides an embodiment of a multi-mode predictive control system for an unmanned surface ship based on deep learning assistance.

[0112] See also Figure 5 An embodiment of the present invention provides a multi-mode predictive control system for an unmanned surface ship assisted by deep learning, which is used to implement the multi-mode predictive control method for an unmanned surface ship assisted by deep learning in the above-mentioned embodiment. The multi-mode predictive control system includes an establishment module 210, a first construction module 211, a sampling combination module 212, a second construction module 213, a migration update module 214, a calculation module 215 and an implementation module 216.

[0113] In this embodiment, the establishing module 210 is used to establish a nonlinear motion model of the surface unmanned ship; wherein the unmanned ship includes multiple operation modes for different task types, and each operation mode has multiple sub-modes.

[0114] In this embodiment, the first construction module 211 is used to construct a preset nonlinear model predictive controller for the surface unmanned ship according to the actual application scenario.

[0115] In this embodiment, the sampling combination module 212 is used to add corresponding obstacle avoidance constraints, and to record the system status and control input data of the surface unmanned vessel in real time during a simulation process with environmental disturbances or measurement errors to obtain a training data set.

[0116] In this embodiment, the second construction module 213 is used to use the training data set to train the LSTM network through a preset existing deep learning training algorithm, and obtain a learning model after the training is completed.

[0117] In this embodiment, the update module 214 is used to obtain the predicted state of the unmanned surface vessel using the learning model and the nonlinear model predictive controller during the operation phase, and quantify the state deviation.

[0118] In this embodiment, the calculation module 215 is used to input the state deviation into a preset nonlinear model predictive controller, obtain a corrected single-step optimal control input by solving the single-step optimal control problem, and drive the unmanned surface vessel.

[0119] In this embodiment, the implementation module 216 is used to repeatedly call the above modules and perform multi-mode predictive control of the surface unmanned vessel through a deep learning-assisted control model.

[0120] The implementation process of the functions and effects of each module in the above-mentioned multi-mode predictive control system is specifically described in the implementation process of the corresponding steps in the above-mentioned method, which will not be repeated here.

[0121] For the multi-mode predictive control system embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The system embodiment described above is only schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present application. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0122] Accordingly, an embodiment of the present invention further provides an electronic device, including one or more processors and a memory, wherein the memory is coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the multi-mode predictive control method for an unmanned ship based on soft switching in the above embodiment. Figure 6 As shown, it is a hardware structure diagram of an electronic device provided by an embodiment of the present invention, except Figure 6 In addition to the processor, memory and network interface shown, any device with data processing capability in which the system in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capability, which will not be described in detail.

[0123] Correspondingly, an embodiment of the present invention also provides a computer-readable storage medium on which computer instructions are stored. When the instructions are executed by a processor, the multi-mode predictive control method of the surface unmanned ship based on deep learning assistance in the above embodiment is implemented.

[0124] The computer-readable storage medium may be an internal storage unit of any device with data processing capability described in any of the aforementioned embodiments, such as a hard disk or a memory. The computer-readable storage medium may also be an external storage device, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capability and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capability, and may also be used to temporarily store data that has been output or is to be output.

[0125] Those skilled in the art will readily appreciate other embodiments of the present application after considering the description and practicing the contents disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The description and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0126] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A multi-mode predictive control method for unmanned surface vessels based on deep learning assistance, It is characterized in that include: Establishing a nonlinear motion model of a surface unmanned ship; wherein the unmanned ship includes multiple operation modes for different mission types, and each operation mode has multiple sub-modes; Constructing a preset nonlinear model predictive controller for the unmanned surface ship; Add obstacle avoidance constraints, and record the system status and control input data of the surface unmanned ship in real time during the simulation process with environmental disturbances or measurement errors to obtain a training data set; Using the training data set to train the long short-term memory network through a preset deep learning training algorithm, and obtaining a learning model after the training is completed; Based on the learning model and the nonlinear model predictive controller, the learning model is used in the operation phase to obtain the predicted state of the surface unmanned ship and quantify the state deviation; The state deviation is input into the preset nonlinear model predictive controller, and the corrected single-step optimal control input is obtained by solving the single-step optimal control problem, and the surface unmanned vessel is driven; Repeat the above process and perform multi-mode predictive control of the surface unmanned vessel through a deep learning-assisted control model.

2. The multi-mode predictive control method for unmanned surface vessels based on deep learning assistance according to claim 1, It is characterized in that The nonlinear motion model of the surface unmanned ship is established; wherein the unmanned ship includes multiple operation modes for different task types, and each operation mode has multiple sub-modes, specifically including: A fixed frame coordinate system tangent to the earth's surface and a motion coordinate system fixed on the hull are established to represent the kinematic model and dynamic model of the surface unmanned ship, which is the nonlinear motion model of the surface unmanned ship. According to different mission types, the control mode of the unmanned ship is divided into four operating modes: straight sailing, turning, point stabilization and trajectory tracking. On the basis of the above four operating modes, each operating mode is subdivided into two modes: speed and collision. Among them, the speed mode includes three types: low speed, medium speed and high speed. The collision mode includes four types: static obstacle, overtaking, encounter and cross encounter according to the obstacle type and the encounter situation of the ship.

3. The multi-mode predictive control method for unmanned surface vessels based on deep learning assistance according to claim 1 or 2, It is characterized in that The nonlinear motion model of the unmanned surface ship includes a kinematic model and a dynamic model, and the expressions thereof are: Among them, M 1 ,M 2 ,M 3 and D 1 ,D 2 ,D 3 are the diagonal elements of the inertia matrix and the diagonal linear components of the damping matrix respectively; x, y, ψ represent the x-axis position, y-axis position and direction in the fixed coordinate system respectively; Respectively represent the instantaneous changes in the x-axis position, y-axis position and direction in the fixed coordinate system; u, v, r represent the x-axis velocity, y-axis velocity and angular velocity in the moving coordinate system; Respectively represent the instantaneous changes of x-axis velocity, y-axis velocity and angular velocity in the motion coordinate system; τ u and τ r They represent the force and torque provided by the two propellers respectively.

4. The multi-mode predictive control method for an unmanned surface ship based on deep learning assistance according to claim 1, It is characterized in that The step of constructing a preset nonlinear model predictive controller for the unmanned surface ship specifically includes: Based on the nonlinear system, a nonlinear model predictive controller is constructed for the surface unmanned ship. The nonlinear model predictive controller solves the optimal control problem in the future finite prediction time domain at each moment at the cost of the deviation between the system state and the reference trajectory and the control input increment: Among them, J is the cost function, is the deviation between the system state and the reference trajectory, x k is the system state variable, is the reference target or trajectory, k represents the kth moment, k+i represents the k+ith moment, f(·) represents the nonlinear system, x min ,x max is the constraint condition of the system state variables, u min ,u max is the system control variable constraint, H represents the finite prediction horizon, represents the quadratic norm x T Qx, represents the quadratic norm x T Rx, Q≥0, R≥0 is an adjustable weight matrix.

5. The multi-mode predictive control method for unmanned surface vessels based on deep learning assistance according to claim 1, It is characterized in that The obstacle avoidance constraint is added by recording the system state and control input data of the surface unmanned ship in real time during a simulation process with environmental disturbances or measurement errors to obtain a training data set, specifically including: Based on the preset nonlinear model predictive controller constructed for the surface unmanned vessel, the corresponding obstacle avoidance constraints are added to the nonlinear model predictive control; In the simulation process, environmental disturbances or measurement errors are added to obtain and record the system state and control input data of the surface unmanned ship in real time to obtain a first data set; The first data set is preprocessed to obtain a training data set; wherein the preprocessing includes importing, extracting features and labels, dividing the data set, normalizing and packaging the data.

6. The multi-mode predictive control method for an unmanned surface ship based on deep learning assistance according to claim 1, It is characterized in that Based on the learning model and the nonlinear model predictive controller, the operation phase uses the learning model to obtain the predicted state of the surface unmanned ship and quantify the state deviation, which specifically includes: Based on the trained learning model, the predicted state of the surface unmanned vessel is obtained according to the historical system state, historical control input data and the current system state and current control input data; According to the predicted state and actual state of the surface unmanned ship, the state deviation of the surface unmanned ship, that is, the state correction data, is calculated.

7. The multi-mode predictive control method for an unmanned surface ship based on deep learning assistance according to claim 1, It is characterized in that The state deviation is input into a preset nonlinear model predictive controller, and a corrected single-step optimal control input is obtained by solving a single-step optimal control problem, and the unmanned ship is driven, specifically including: The state correction data is input into the preset nonlinear model predictive controller, and the single-step optimal control problem is solved by adjusting the system state and control input weight of the surface unmanned ship, thereby calculating the optimal control input data at the current moment; Applying the optimal control input data to the unmanned surface ship to drive the unmanned surface ship to move according to the revised control strategy; In each control cycle, the corrected control input data and the real-time system status of the surface unmanned vessel are recorded.

8. A deep learning-assisted multi-mode predictive control system for an unmanned surface ship, used to implement the deep learning-assisted multi-mode predictive control method for an unmanned surface ship according to any one of claims 1 to 7, It is characterized in that include: Establishing a module for establishing a nonlinear motion model of a surface unmanned ship; wherein the unmanned ship includes multiple operation modes for different task types, and each operation mode has multiple sub-modes; A first construction module is used to construct a preset nonlinear model predictive controller for the unmanned surface ship; The sampling combination module is used to add obstacle avoidance constraints and record the system status and control input data of the surface unmanned ship in real time during the simulation process with environmental disturbances or measurement errors to obtain the training data set; A second construction module is used to train the long short-term memory network using the training data set through a preset deep learning training algorithm, and obtain a learning model after the training is completed; An update module is introduced, which is used to obtain the predicted state of the surface unmanned ship by using the learning model and the nonlinear model predictive controller during the operation phase, and quantify the state deviation; A calculation module, used for inputting the state deviation into a preset nonlinear model predictive controller, obtaining a corrected single-step optimal control input by solving the single-step optimal control problem, and driving the surface unmanned vessel; and An implementation module is used to repeatedly call the above modules and perform multi-mode predictive control of surface unmanned ships through a deep learning-assisted control model.

9. An electronic device, It is characterized in that include: one or more processors; and a memory coupled to the processor; Wherein, the memory is used to store program data, and the processor is used to execute the program data to implement the multi-mode predictive control method for surface unmanned ships based on deep learning assistance as described in any one of claims 1-7.

10. A computer-readable storage medium, It is characterized in that Computer instructions are stored thereon, which, when executed by a processor, are used to implement the multi-mode predictive control method for an unmanned surface ship based on deep learning assistance as described in any one of claims 1-7.

Images