Inland ship safety multi-task optimal tracking control method

Abstract

The invention discloses an inland ship safety multi-task optimal tracking control method, and belongs to the technical field of inland ship tracking control. The method comprises the following steps: establishing discrete kinematics and dynamics equations of an inland ship, and decoupling a ship nonlinear system into a kinematics subsystem and a dynamics subsystem; in the kinematics subsystem, an optimal virtual control law meeting the kinematics safety multi-task requirement is designed; and based on the optimal virtual control law, in the dynamics subsystem, designing an optimal control law meeting dynamics safety multi-task requirements, and guiding the inland ship to track a preset trajectory. According to the method, the expected trajectory of the inland ship can be tracked in a safer, more accurate and more energy-saving manner, and the safe multi-task optimal tracking performance of the inland ship is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of inland river vessel tracking control, and in particular to a safe multi-task optimal tracking control method for inland river vessels. Background Art

[0002] Inland waterway navigation is a complex and dynamic environment characterized by narrow and winding waterways, high vessel density, and dynamically changing water flow conditions. Navigation safety is paramount and must take precedence over other performance metrics, such as tracking error and control cost. Approximate dynamic programming methods are widely used to achieve optimal ship control, with the Hamilton-Jacobi-Bellman equation being a mainstream approach. Single-layer neural networks, typically based on the AC learning framework and using basis functions as activation functions, are used to approximate value functions and control inputs. Using single-layer neural networks to approximate nonlinear systems requires either appropriate selection of basis functions or the use of a large number of hidden layer neurons to randomly select inputs for hidden layer weights, which presents certain limitations. Multilayer neural networks can alleviate this need for basis function selection and require fewer neurons per layer to achieve the same level of accuracy compared to single-layer neural networks. However, weight adjustment remains challenging. At the same time, existing ship optimal control strategies usually consider single-objective issues such as tracking error and control cost, without taking safety into consideration. When inland ships sail in narrow waters, their navigation safety is extremely important. The development of a multi-task optimal tracking control method with high safety, high-precision tracking capability and low consumption cost has become a key issue that needs to be urgently solved in the current field of inland ship tracking control.

[0003] Existing AC learning-based control strategies, when used in multi-task scenarios, typically require retraining on cross-task data to adapt to environmental changes. However, when using online learning, neural networks often suffer from "catastrophic forgetting" during training for new tasks. This occurs when knowledge from previous tasks is overwritten by the new ones, resulting in degradation of control performance in previously learned scenarios. This limitation severely restricts the practical application of optimal control theory in complex inland river environments. Summary of the Invention

[0004] In response to the deficiencies in the above-mentioned prior art, the present invention provides a safe multi-task optimal tracking control method for inland vessels, which solves the problem of catastrophic forgetting of traditional learning strategies in a multi-task environment considering high safety, high-precision tracking capability and low consumption cost, realizes the safe and efficient navigation of inland vessels in narrow waters, and improves the trajectory tracking performance of inland vessels.

[0005] The first aspect of the present invention provides a safe multi-task optimal tracking control method for an inland waterway vessel, the method comprising: Establish discrete kinematic and dynamic equations for inland waterway vessels and decouple the nonlinear system of the vessel into kinematic and dynamic subsystems; In the kinematic subsystem, an optimal virtual control law is designed to meet the kinematic safety multi-task requirements, including: Combined with the preset trajectory, the kinematic level trajectory tracking error is generated; Define a navigation area set and design a kinematic level ship navigation obstacle function based on the navigation area set; Based on the kinematic-level trajectory tracking error, the control commands of the kinematic subsystem, and the kinematic-level ship navigation obstacle function, a kinematic multi-task objective cost function with safety dynamic constraints is designed. The Hamilton-Jacobi-Bellman equation is constructed, and combined with the back-recursion technique and adaptive dynamic programming algorithm, an ideal virtual control law that meets the optimality requirements is solved. Design a kinematics-level action-evaluation multi-layer neural network architecture to approximate the kinematic optimal multi-task objective cost function and the unknown uncertainties of the ideal virtual control law, obtain the optimal virtual control law, and update the kinematics-level action-evaluation multi-layer neural network weights; Based on the optimal virtual control law, an optimal control law that meets the multi-task requirements of dynamic safety is designed in the dynamic subsystem to guide inland waterway vessels to follow the preset trajectory. Specifically, the following are the steps: Calculate the speed tracking error at the dynamic level based on the optimal virtual control law and ship speed information; Define a set of navigation speed regions and design a dynamic level ship speed obstacle function based on the set of navigation speed regions; Based on the dynamic-level speed tracking error, the control commands of the dynamic subsystem, and the dynamic-level ship speed obstacle function, a dynamic multi-task objective cost function with dynamic speed safety constraints is designed. The Hamilton-Jacobi-Bellman equation is constructed and solved using an adaptive dynamic programming algorithm to obtain the ideal control law that meets the optimality requirements. A dynamic hierarchical action-evaluation multi-layer neural network architecture is designed to approximate the dynamic optimal multi-task objective cost function and the unknown uncertainties of the ideal control law, obtain the optimal control law, guide the inland vessel to track the preset trajectory, and update the dynamic hierarchical action-evaluation multi-layer neural network weights.

[0006] In the above solution, the navigation area set includes an absolutely safe navigation area, a relatively safe navigation area and a dangerous navigation area.

[0007] In the above solution, the design of the kinematic level ship navigation obstacle function based on the navigation area set includes: Determine the navigation area set to which the kinematic level trajectory tracking error belongs: When the kinematic level trajectory tracking error is in the absolutely safe navigation area, a logarithmic obstacle function is designed; When the kinematic-level trajectory tracking error is in a relatively safe navigation area, a composite obstacle function is constructed, the composite obstacle function including a danger repelling function and a safety attracting function, to guide the kinematic-level trajectory tracking error toward the origin and away from the navigation danger boundary; wherein the navigation danger boundary is the boundary between the relatively safe navigation area and the dangerous navigation area; When the kinematic level trajectory tracking error is located in a dangerous navigation area, the barrier function value tends to infinity.

[0008] In the above scheme, the kinematic multi-task objective cost function with safety dynamic constraints includes a trajectory tracking error penalty function based on the kinematic level trajectory tracking error, a control energy consumption function based on the control command of the kinematic subsystem, and a safety cost function based on the kinematic level ship navigation obstacle function. At the same time, the kinematic level trajectory tracking error satisfies the inland waterway ship safety dynamic constraints, that is, the kinematic level trajectory tracking error is located in the navigation safety set, which includes an absolutely safe navigation area and a relatively safe navigation area. The updating of kinematic level actions - evaluating multi-layer neural network weights, includes: The normalized gradient descent method is used to derive the weight update law of the kinematic-level action multi-layer neural network and the weight update law of the kinematic-level evaluation multi-layer neural network. The lifelong learning method and the online elastic weight consolidation method are adopted in the weight update law of the kinematic-level action multi-layer neural network and the weight update law of the kinematic-level evaluation multi-layer neural network. The regularization term is introduced and combined with the information diagonal matrix of historical tasks to suppress the catastrophic forgetting of the control strategy in multi-task scenarios. At the same time, the kinematic control obstacle function is integrated as a safety constraint condition in the kinematic-level action multi-layer neural network update law, which cooperates with the kinematic-level ship navigation obstacle function to keep the ship trajectory within the navigation safety set.

[0009] In the above solution, the navigation speed area set includes an absolutely safe navigation speed area, a relatively safe navigation speed area and a dangerous navigation speed area.

[0010] In the above solution, the design of the dynamic level ship speed obstacle function based on the navigation speed region set includes: Determine the navigation speed region set to which the dynamic level speed tracking error belongs: When the dynamic level speed tracking error is in the absolutely safe area of ​​navigation speed, a logarithmic barrier function is designed; When the dynamic level speed tracking error is in a relatively safe navigation speed region, a composite obstacle function is constructed, the composite obstacle function including a danger repelling function and a safety attracting function, guiding the dynamic level speed tracking error toward the origin and away from the navigation speed danger boundary; wherein the navigation speed danger boundary is the boundary between the relatively safe navigation speed region and the navigation speed danger region; When the dynamic level speed tracking error is in the navigation speed danger zone, the barrier function value tends to infinity.

[0011] In the above scheme, the dynamic multi-task objective cost function with dynamic speed safety constraints includes a speed tracking error penalty function of the dynamic level speed tracking error, a control energy consumption function based on the control command of the dynamic subsystem, and a safety cost function based on the dynamic level ship speed obstacle function. At the same time, the dynamic level speed tracking error satisfies the dynamic safety constraints of inland ships, that is, the dynamic level speed tracking error is located in the navigation speed safety set, which includes an absolute navigation speed safety area and a relatively safe navigation speed area. The updating dynamics level action - evaluating the weights of the multi-layer neural network, includes: Using the normalized gradient descent method, the weight update law of the dynamic-level action multi-layer neural network and the weight update law of the dynamic-level evaluation multi-layer neural network are derived. The lifelong learning method and the online elastic weight consolidation method are adopted in the weight update law of the dynamic-level action multi-layer neural network and the weight update law of the dynamic-level evaluation multi-layer neural network. The regularization term is introduced and combined with the information diagonal matrix of the historical tasks to suppress the catastrophic forgetting of the control strategy in the multi-task scenario; at the same time, the dynamic control obstacle function is integrated as a safety constraint condition in the dynamic-level action multi-layer neural network update law, and cooperates with the dynamic-level ship speed obstacle function to keep the ship speed within the safe set of navigation speeds.

[0012] According to a second aspect of the present invention, there is provided a safe multi-task optimal tracking control system for an inland waterway vessel, the system applying the safe multi-task optimal tracking control method for an inland waterway vessel according to any one of the first aspects, comprising: Model building and decoupling module, used to establish discrete kinematic and dynamic equations for inland waterway vessels, and decouple the nonlinear system of the vessel into kinematic and dynamic subsystems; The kinematic subsystem control module, including an online learning unit and a safety decision-making unit, generates the optimal virtual control law by minimizing the kinematic multi-task objective cost function with safety dynamic constraints; The dynamics subsystem control module, including an online learning unit and a safety decision-making unit, generates an optimal control law by minimizing the dynamics multi-task objective cost function with speed safety dynamic constraints, and guides the inland waterway vessel to follow the preset trajectory; The online learning unit uses a corresponding level of action-evaluation multi-layer neural network architecture, lifelong learning methods, and elastic weight consolidation algorithms to update network weights in real time. The safety decision-making unit monitors the tracking error of the corresponding level and the distance to the danger zone in real time, and updates the obstacle function of the corresponding level. The danger zone is reflected as the position of the obstacle at the kinematic level, that is, the area that the ship cannot reach, and is reflected as the limit of the ship's navigation speed at the dynamic level.

[0013] According to a third aspect of the present invention, a computer device is provided, comprising: a processor and a memory, wherein the memory stores programs or instructions that can be run on the processor, and when the programs or instructions are executed by the processor, the steps of the inland vessel safety multi-task optimal tracking control method described in any one of the first aspects are implemented.

[0014] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the inland vessel safety multi-task optimal tracking control method described in any one of the first aspects are implemented.

[0015] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art: This invention defines safety sets and obstacle functions at the kinematic and dynamic levels. By adjusting the obstacle functions based on the actual conditions of ships of different sizes and in different navigation waters, the danger set boundaries are adjusted in real time, resolving the issue of static boundaries being unable to adapt to sudden risks. A multi-task objective cost function with safety and dynamic constraints is designed to achieve safer, more accurate, and more energy-efficient multi-task optimal tracking performance for inland vessels. A lifelong learning-based action-evaluation multi-layer neural network architecture overcomes the vanishing gradient problem of traditional multi-layer neural networks and achieves higher nonlinear approximation accuracy compared to fixed basis function methods using single-layer neural networks. Furthermore, this approach addresses the challenges of multi-task continuous learning and anti-forgetting performance for ship trajectory tracking in dynamic and complex environments. The action-evaluation multi-layer neural network and elastic weight curing technology enable the coexistence of new and old knowledge, addressing the "catastrophic forgetting" problem of traditional AC learning and enabling lifelong accumulation and efficient reuse of multi-scenario strategies. The proposed safe multi-task optimal tracking control method for inland vessels addresses the catastrophic forgetting problem of traditional AC learning strategies in multi-task environments requiring high safety, high-precision tracking capabilities, and low consumption costs. This invention enables safe and efficient navigation of inland vessels. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flow chart of a safe multi-task optimal tracking control method for inland waterway vessels provided in an embodiment of the present application; Figure 2 A schematic diagram of a safe multi-task optimal tracking control method for inland waterway vessels provided in an embodiment of the present application; Figure 3 A trajectory tracking effect diagram provided in an embodiment of the present application; Figure 4 An XY plane tracking effect diagram provided by an embodiment of the present application; Figure 5 A total cost map provided for an embodiment of the present application; Figure 6 A schematic diagram of the structure of a multi-task optimal tracking control system for inland waterway vessel safety provided in an embodiment of the present application; Figure 7 A schematic diagram of the hardware structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and examples. 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. Based on the embodiments provided in this application, all other embodiments obtained by those of ordinary skill in the art without making creative work are within the scope of protection of the present invention.

[0018] Obviously, the drawings described below are merely examples or embodiments of the present application. Those skilled in the art can, without inventive effort, apply the present application to other similar scenarios based on these drawings. Furthermore, it is also understood that, although the effort involved in such a development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, changes in design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as an insufficiency of the content disclosed in this application.

[0019] References to "embodiments" in this application mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described in this application may be combined with other embodiments unless there is a conflict.

[0020] Unless otherwise defined, technical or scientific terms used herein shall have the ordinary meaning as understood by persons of ordinary skill in the art to which this application belongs. The terms "a," "an," "an," "the," and similar expressions used herein do not denote quantitative limitations and may refer to either the singular or the plural. The terms "comprise," "include," "have," and any variations thereof, used herein, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or modules (units) is not limited to the listed steps or units but may also include steps or units not listed, or may include other steps or units inherent to the process, method, product, or apparatus. The terms "connected," "connected," "coupled," and similar expressions used herein are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. As used herein, "plurality" means two or more. "And / or" describes an association between associated objects, indicating that three possible relationships exist. For example, "A and / or B" may mean: A exists alone; A and B exist simultaneously; or B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific order for the objects.

[0021] This application provides a multi-task optimal tracking control method for inland waterway vessel safety, aiming to achieve the optimal performance of multiple objectives such as high inland waterway vessel tracking accuracy, navigation safety, and low control cost. Figure 1 and Figure 2 As shown, the specific process of this method is: Step 1: Establish the discrete kinematic and dynamic equations of the inland waterway ship and decouple the ship's nonlinear system into kinematic and dynamic subsystems; Step 2: In the kinematic subsystem, design the optimal virtual control law that meets the kinematic safety multi-task requirements; Step 3: Based on the optimal virtual control law, in the dynamic subsystem, design the optimal control law that meets the dynamic safety multi-task requirements to guide the inland vessel to track the preset trajectory.

[0022] In some embodiments, the specific steps of designing an optimal virtual control law that meets the kinematic safety multi-tasking requirements include: Combined with the preset trajectory, the kinematic level trajectory tracking error is generated; Define the navigation area set and design the kinematic level ship navigation obstacle function; Design a multi-task objective cost function with safety dynamic constraints at the kinematic level; construct the Hamilton-Jacobi-Bellman equation, and combine the back-recursion technique with an adaptive dynamic programming algorithm to solve the ideal virtual control law that meets the optimality requirements; A kinematic hierarchical action-evaluation multi-layer neural network architecture is designed to approximate the kinematic optimal multi-task objective cost function and the unknown uncertainties of the ideal virtual control law, obtain the optimal virtual control law, and update the kinematic hierarchical action-evaluation multi-layer neural network weights online.

[0023] In some embodiments, the specific steps of designing an optimal control law that meets the dynamic safety multi-task requirements include: Calculate the speed tracking error at the dynamic level based on the optimal virtual control law and ship speed information; Design speed obstacle function for inland waterway vessels; Design a dynamic multi-task objective cost function with dynamic constraints on speed safety; construct the Hamilton-Jacobi-Bellman equation and solve it based on an adaptive dynamic programming algorithm to achieve the ideal control law that meets the optimality requirements; A dynamic hierarchical action-evaluation multi-layer neural network architecture is designed to approximate the dynamic optimal multi-task objective cost function and the unknown uncertainties of the ideal control law, obtain the optimal control law, guide inland vessels to track the preset trajectory, and update the dynamic hierarchical action-evaluation multi-layer neural network weights online.

[0024] In some embodiments, designing a kinematic-level ship navigation obstacle function includes: Determine whether the kinematic level trajectory tracking error belongs to the safe navigation area set; When the kinematic level trajectory tracking error is within the safe navigation area set, a logarithmic obstacle function is designed; When the kinematic level trajectory tracking error is outside the safe navigation area set but does not enter the danger area set, a composite obstacle function is constructed to guide the error toward the origin and away from the danger set boundary; the composite obstacle function includes a danger repulsion function and a safety attraction function.

[0025] In some embodiments, a multi-task objective cost function with safety dynamic constraints at the kinematic level includes a tracking error penalty function, a control energy consumption function, and a safety cost function. The safety cost function adjusts the penalty weight of the obstacle function to adaptively match safety priorities in different task scenarios.

[0026] In some embodiments, updating the action-evaluation multi-layer neural network weights specifically includes: adopting a lifelong learning method and an online elastic weight consolidation method in the action and evaluation multi-layer neural network weight update law, introducing a regularization term, and combining the information diagonal matrix of historical tasks to suppress catastrophic forgetting of the control strategy in a multi-task scenario.

[0027] Updating actions and evaluating multi-layer neural network weights also includes: integrating a kinematic control obstacle function into a kinematic level action multi-layer neural network update law, and coordinating with a kinematic level ship navigation obstacle function to keep the trajectory within a navigation safety set, even if the kinematic level trajectory tracking error moves from a dangerous navigation area to a relatively safe navigation area, and then to an absolutely safe navigation area.

[0028] The dynamics subsystem is basically the same as the kinematics subsystem, the difference is that the kinematics subsystem is trajectory tracking and the dynamics subsystem is velocity tracking.

[0029] Specifically, such as Figure 1 and Figure 2 As shown, the inland waterway vessel safety multi-task optimal tracking control method of the embodiment of the present application is specifically implemented as follows: Step 1: Establish the discrete kinematic and dynamic equations of the inland waterway ship and decouple the ship's nonlinear system into kinematic and dynamic subsystems:

[0030] in, and The navigation status of the ship at time k and time k+1 respectively, and the ship position in the geodetic coordinate system is and heading constitute; is the transformation matrix; and is the velocity state of the ship at the kth moment and the k+1th moment in the hull coordinate, which is composed of the ship's longitudinal and transverse velocities and angular velocity; is the inertia matrix; is the Coliolis centripetal matrix; is the damping matrix; The dynamic level control input consists of the ship's surge control force, sway control force, and yaw control moment. The actual control input is the optimal control input designed below. ; External interference is composed of wind, waves and currents in the ocean environment, which can be expressed as ,in is a first-order Markov process, is a zero-mean Gaussian white noise process, and is a diagonal constant matrix, It is a time-varying function of the sine and cosine combination. In practice, external interference is caused by wind and waves, and there is no specific formula to express it.

[0031] Step 2: In the kinematic subsystem, design an approximately optimal virtual control law that meets the requirements of safe multi-tasking. The specific steps include: (1) Combine the preset trajectory to generate the kinematic level trajectory tracking error;

[0032] in, is the kinematic level trajectory tracking error, The actual navigation status. is the expected trajectory.

[0033] (2) Define a set of navigation areas and design a kinematic level ship navigation obstacle function; the navigation area set includes an absolutely safe navigation area , relatively safe navigation areas and navigation hazards areas .

[0034] The absolutely safe navigation area The tracking error is less than 0.15 times the ship width, which is a dangerous navigation area. The tracking error is greater than 0.5 times the ship width, which is a relatively safe navigation area. The tracking error is greater than or equal to 0.15 times and less than or equal to 0.5 times the ship's width. It should be noted that when high-precision ship tracking is required, users can preset the navigation area based on actual conditions.

[0035] Based on the inland river ship navigation area set, the kinematic level ship navigation obstacle function corresponding to the inland river ship is designed as follows:

[0036] Among them, the trajectory tracking error Depend on Composition, corresponding to the ship position and heading ; is the kinematic level ship navigation obstacle function, Tracking error No. i Status The corresponding navigation obstacle function; and The trajectory tracking error The upper and lower limits of the absolute safe navigation area are determined by and constitute, and Tracking error No. i Status The corresponding upper and lower limits of the absolutely safe navigation area; It is a relatively safe navigation area; It is a dangerous area for navigation; It is the boundary between the relatively safe navigation area and the navigation risk area, that is, the navigation risk boundary; and are the distance functionals of the tracking error to the origin and the boundary of the danger zone, respectively.

[0037] It should be noted that when the tracking error is in the navigation danger area or near the danger boundary, the navigation obstacle function value tends to infinity, guiding the ship away from the danger. To infinity.

[0038] (3) Design a multi-task objective cost function with safety dynamic constraints at the kinematic level; construct the Hamilton-Jacobi-Bellman equation, combine the back-recursion technique with the adaptive dynamic programming algorithm, and solve the ideal virtual control law that meets the optimality requirements.

[0039] Based on the kinematic level tracking error and the ship navigation obstacle function, the multi-task objective cost function with safety dynamic constraints is constructed as follows:

[0040] in, For the k The time is a multi-task objective cost function with safety dynamic constraints at the kinematic level, Satisfy the dynamic constraints of inland waterway ship safety, that is, the kinematic level tracking error must be satisfied Mustering for safety in navigation; For the k Control commands for the kinematic subsystem at each moment, 、 、 are the penalty terms for tracking error, control command, and ship navigation obstacle function in the kinematic subsystem respectively.

[0041] The navigation safety set includes the absolutely safe navigation area and the relatively safe navigation area. The multi-task objective cost function is composed of tracking error cost, control input cost, and safety cost.

[0042] For the safe optimal tracking control of the inland waterway vessel operation subsystem, it is necessary to find the kinematic ideal virtual control command to meet the dynamic constraints of the inland waterway vessel safety while minimizing the multi-task objective cost function, namely:

[0043] Where, is the optimal multi-task objective cost function at the kth moment.

[0044] To solve the above function, the Hamilton-Jacobi-Bellman equation is constructed as:

[0045] Where, is the discount factor; minimize , solve , the ideal virtual control law is derived:

[0046] Where, for The optimal multi-task objective cost function at the k+1th moment with respect to the tracking error at the k+1th moment The partial derivative of .

[0047] (4) Design a kinematic hierarchical action-evaluation multi-layer neural network architecture to approximate the optimal multi-task objective cost function and the unknown uncertainties of the ideal virtual control law, obtain the optimal virtual control law, and update the action-evaluation multi-layer neural network weights online.

[0048] Use evaluation multi-layer neural network to approximate the optimal multi-task objective cost function:

[0049] in, is the basis function, Evaluate multi-layer neural network activation functions for the kinematic level, and Evaluate the target weights corresponding to activation functions and basis functions in multi-layer neural networks for the kinematic level; To evaluate the neural network modeling error.

[0050] The optimal multi-task objective cost function estimate is obtained based on the evaluation neural network:

[0051] Where, and To evaluate the actual weights corresponding to the activation function and basis function in the multi-layer neural network.

[0052] The optimal virtual control law is obtained based on the action neural network: ; Where, and is the actual weight corresponding to the activation function and basis function in the action multi-layer neural network; and are the basis functions and activation functions of the multi-layer neural network for kinematic actions.

[0053] The kinematic time difference error is calculated as:

[0054] Where, For the k -1 moment kinematic tracking error.

[0055] Based on the ideal virtual control law and the optimal virtual control law, the virtual control law error is calculated as:

[0056] Where, To evaluate the activation function output in a multi-layer neural network Find the partial derivative.

[0057] In the action-evaluation multi-layer neural network architecture of the kinematic subsystem, lifelong learning methods and elastic weight consolidation techniques are used. It is important to note that a new diagonal matrix is ​​used here to reflect the importance of the connections in the neural network, replacing the original Fisher matrix, thereby avoiding the complexity of calculating the Fisher matrix. To describe each task The importance of task information is defined as , , , , the diagonal elements in the matrix are:

[0058] and

[0059] Where, , , , For the task The current weight of , , , are the connection signals corresponding to the weights respectively.

[0060] It should be noted that each task in the kinematic system The corresponding navigation trajectory is caused by the change of actual operation requirements. Specifically, it can be divided into straight navigation, circular trajectory, sine-cosine combination curve, straight line + circle combination curve, 8-shaped trajectory, etc., which can be defined by the user.

[0061] Constructing action multi-layer neural networks and evaluating the performance metrics of multi-layer neural networks:

[0062] Where, , , , It's a task The optimal weight value for training; , , , For the task The current weight of , , , is the regularization coefficient, which controls the balance of importance between new and old tasks. The larger the value, the stronger the retention of the old task weights. is the virtual control law error; is the kinematic timing difference error.

[0063] Using the normalized gradient descent method, the weight update law of the evaluation multi-layer neural network adapted to lifelong learning is derived as follows:

[0064] And the action multi-layer neural network weight update law is:

[0065] Where, 、 、 、 is the weight of the evaluation network and action network at the kth moment; 、 、 、 The weights of the evaluation network and action network at the k+1th moment; 、 、 、 is the design parameter; , , , is the regularization coefficient; and Vector for design; ; , , , is the ideal weight matrix at the end of task j; is the optimal virtual control input; and To design safety parameters, is the judgment operator, which is defined as:

[0066] in, is the ship navigation obstacle function, that is ; Indicates the optimal virtual control input The amount of change in the control barrier function under action; is the design parameter.

[0067] It should be noted that, in the optimal virtual control input Under the action of the control obstacle function, if the change of the ship's navigation obstacle function satisfies the control obstacle function constraint, the judgment factor is 0; otherwise, the judgment operator is 1, and the weight update law of the action multi-layer neural network is adjusted by the navigation obstacle function.

[0068] Step 3: Based on the optimal virtual control law, design an optimal control law that meets the safety multi-task requirements in the dynamic subsystem to guide the inland waterway vessel to follow the preset trajectory. The specific steps include: (1) Based on the optimal virtual control law and ship speed information, calculate the dynamic tracking error:

[0069] Where, For the k Time velocity tracking error; For the k The velocity state of the ship in the hull coordinate at the moment is composed of the ship's longitudinal and transverse speeds and angular velocity; is the optimal virtual control law.

[0070] (2) Define a set of navigation speed regions and design a dynamic level ship speed obstacle function; the navigation speed region set includes an absolute speed safety region , relatively safe speed area and speed danger zones .

[0071] The absolute safety area of ​​the speed The speed tracking error is less than 0.15 times the sailing speed, which is the speed danger zone. The speed tracking error is greater than 0.5 times the navigation speed, which is a relatively safe speed area. The tracking error is greater than or equal to 0.15 times and less than or equal to 0.5 times the sailing speed. It should be noted that the user can preset the speed zone set according to the actual maneuverability of the ship.

[0072] Based on the navigation speed region set, the dynamic level ship speed obstacle function is designed as follows:

[0073] Among them, the navigation speed error Depend on composition, corresponding to the ship's longitudinal and transverse speeds and angular velocity respectively; is the ship speed obstacle function at the dynamic level, is the navigation speed error No. i Status The corresponding navigation obstacle function; and Sailing speed error The upper and lower limits of the absolute safe navigation area are determined by and constitute, and is the navigation speed error No. i Status The corresponding upper and lower limits of the absolute safety zone for navigation speed; It is a relatively safe navigation area; It is a speed danger zone; It is the critical point between the relatively safe navigation area and the dangerous speed area, that is, the dangerous speed boundary; and are the distance functionals from the speed error to the origin and the boundary of the danger zone, respectively.

[0074] (3) Design a dynamic multi-task objective cost function with dynamic constraints on speed safety; construct the Hamilton-Jacobi-Bellman equation and combine it with an adaptive dynamic programming algorithm to solve the ideal control law that meets the optimality requirements.

[0075] Based on the dynamic tracking error and speed obstacle function, the dynamic multi-task objective cost function with speed safety dynamic constraints is constructed as follows:

[0076] in, is the dynamic multi-task objective cost function, For the dynamic constraints of inland waterway ship speed safety, the dynamic tracking error must be within the navigation speed safety set; For the k Control commands for the moment dynamics subsystem, 、 、 are the penalty terms of tracking error, control command, and speed obstacle function in the dynamic subsystem respectively.

[0077] It should be noted that the actual control command of the kinematic subsystem is the optimal virtual control input , corresponds to the optimal navigation speed of the ship in the kinematic subsystem, that is, the ideal speed that enables the ship to track the desired trajectory, which is composed of the surge speed, the sway speed and the yaw angular velocity. The actual control command of the dynamic subsystem is the optimal control input , which corresponds to the optimal control force and torque vector generated by the ship propulsion system in the dynamic subsystem, and the control input consists of the ship's surge control force, sway control force and bow roll control torque.

[0078] The navigation speed safety set includes the navigation speed absolute safety area set and the navigation speed relative safety area set. The dynamic multi-task objective cost function is composed of tracking error cost, control input cost, and speed safety cost.

[0079] For the safe optimal tracking control of the inland vessel dynamics subsystem, it is necessary to find the ideal dynamic control law that satisfies the dynamic constraints of the inland vessel speed while minimizing the dynamic multi-task objective cost function, namely:

[0080] Where, is the dynamic optimal multi-task objective cost function at the kth moment.

[0081] To solve the above function, the Hamilton-Jacobi-Bellman equation is constructed as:

[0082] Where, is the discount factor.

[0083] minimize , solve , derive the ideal control law:

[0084] Where, for The k+1th moment dynamic optimal multi-task objective cost function with respect to the k+1th moment dynamic tracking error The partial derivative of .

[0085] (4) Design a dynamic hierarchical action-evaluation multi-layer neural network architecture to approximate the dynamic optimal multi-task objective cost function and the unknown uncertainties of the ideal control law, obtain the optimal control law, guide the inland waterway vessel to track the preset trajectory, and update the dynamic hierarchical action-evaluation multi-layer neural network weights online; Using an evaluation multi-layer neural network to approximate the dynamics optimal multi-task objective cost function:

[0086] in, and To dynamically evaluate the basis functions and activation functions of multilayer neural networks, and is the target weight corresponding to the activation function and basis function in the dynamic multi-layer neural network; To evaluate the neural network modeling error.

[0087] The optimal multi-task objective cost function estimate is obtained based on the evaluation neural network:

[0088] Where, and The actual weights corresponding to activation functions and basis functions in multi-layer neural networks are evaluated for dynamics.

[0089] The optimal control law is obtained based on the action neural network: ; Where, and is the actual weight corresponding to the activation function and basis function in the multi-layer neural network of dynamic actions; and Basis and activation functions for dynamics evaluation of multilayer neural networks.

[0090] The error of the calculated dynamic time difference is:

[0091] Where, For the k -1 moment dynamic tracking error.

[0092] Based on the ideal control law and the optimal control law, the control law error is calculated as:

[0093] Where, To evaluate the activation function output in a multi-layer neural network Find the partial derivative.

[0094] In the action-evaluation multi-layer neural network architecture of the dynamics subsystem, lifelong learning methods and elastic weight consolidation technology are used to describe each task. The importance of task information is defined as , , , , the diagonal elements in the matrix are:

[0095] and

[0096] Where, , , , For the task The current weight of , , , are the connection signals corresponding to the weights respectively.

[0097] It should be noted that in the dynamic system each task The corresponding navigation speed is the change caused by the navigation trajectory transformation, which corresponds to the navigation trajectory in the kinematic system and can also be defined by the user.

[0098] Constructing action multi-layer neural networks and evaluating the performance metrics of multi-layer neural networks:

[0099] Where, , , , It's a task The optimal weight value for training; , , , For the task The current weight of , , , is the regularization coefficient, which controls the balance of importance between new and old tasks. The larger the value, the stronger the retention of the old task weights. is the control law error; is the dynamic time difference error.

[0100] Using the normalized gradient descent method, the weight update law of the evaluation multi-layer neural network adapted to lifelong learning is derived as follows:

[0101] And the action multi-layer neural network weight update law is:

[0102] Where, 、 、 、 is the weight of the evaluation network and action network at the kth moment; 、 、 、 The weights of the evaluation network and action network at the k+1th moment; 、 、 、 is the design parameter; , , , is the regularization coefficient; and Vector for design; ; , , , For the task j The ideal weight matrix at the end; is the optimal control input; and To design safety parameters, is the judgment operator, which is defined as:

[0103] in, is the barrier function of velocity error, Indicates that the optimal control input The change in the barrier function of the velocity error under action; is the design parameter.

[0104] It should be noted that, in the optimal control input Under the action, if the change of the speed error obstacle function satisfies the control obstacle function constraint, the judgment factor is 0; otherwise, the judgment operator is 1, and the speed obstacle function is used to adjust the action multi-layer neural network weight update law.

[0105] The method of this application has the following effects: Figure 3 、 Figure 4 and Figure 5 As shown, Figure 3 A trajectory tracking effect diagram provided in an embodiment of the present application; Figure 4 An XY plane tracking effect diagram provided by an embodiment of the present application; Figure 5 A total cost diagram is provided for an embodiment of the present application.

[0106] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0107] The present application also provides a multi-task optimal tracking control system for inland waterway vessel safety. These systems are used to implement the above-mentioned embodiments and preferred implementations, and details already described are omitted. Although the systems described in the following embodiments are preferably implemented in software, implementation in hardware or a combination of software and hardware is also possible and contemplated.

[0108] like Figure 6 As shown, the inland waterway vessel safety multi-task optimal tracking control system according to the embodiment of the present application may include: Model construction and decoupling module 201, kinematic subsystem module 202 and dynamic subsystem module 203, the kinematic subsystem module 202 and the dynamic subsystem module 203 both include an online learning unit and a safety decision module; The model building and decoupling module 201 is used to establish discrete kinematic and dynamic equations of the inland waterway vessel and decouple the nonlinear system of the vessel into a kinematic subsystem and a dynamic subsystem; The kinematic subsystem control module 202 includes an online learning unit and a safety decision module. It uses a backward recursive technique and an adaptive dynamic programming algorithm to minimize the multi-task objective cost function with safety dynamic constraints at the kinematic level and generate the optimal virtual control law. The dynamics subsystem control module 203 includes an online learning unit and a safety decision module. By minimizing the multi-task objective cost function with safety dynamic constraints at the dynamic level, it designs an optimal control law that meets the dynamic safety multi-task requirements and guides the inland vessel to follow a preset trajectory.

[0109] The online learning unit uses an action-evaluation multi-layer neural network architecture, lifelong learning methods, and an elastic weight consolidation algorithm to approximate the ideal virtual control law and the ideal control law online and update the multi-layer neural network weights. The safety decision unit monitors the tracking error and the distance to the danger zone in real time and updates the obstacle function. The danger zone is reflected at the kinematic level as the location of obstacles and inaccessible areas for inland vessels, and at the dynamic level as speed restrictions for ships.

[0110] It should be noted that the steps shown in the above process or the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0111] In addition, combined Figure 1 and Figure 2 The inland waterway vessel safety multi-task optimal tracking control method described in the embodiment of the present application can be implemented by a computer device. Figure 7 Schematic diagram of the hardware structure of the computer device of the embodiment of the present application. Figure 7 As shown, the device may include a processor 301 and a memory 302 storing computer program instructions.

[0112] Specifically, the processor 301 may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0113] Memory 302 may include a large-capacity memory for data or instructions. By way of example, and not limitation, memory 302 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk, a magneto-optical disk, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 302 may include removable or non-removable (or fixed) media. Where appropriate, memory 302 may be internal or external to the data processing device. In certain embodiments, memory 302 is non-volatile memory. In certain embodiments, memory 302 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically alterable ROM (EAROM) or a flash memory (FLASH), or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), where the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data out dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0114] The memory 302 may be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 301 .

[0115] The processor 301 reads and executes the computer program instructions stored in the memory 302 to implement any one of the inland vessel safety multi-task optimal tracking control methods in the above embodiments.

[0116] In some embodiments, the inland waterway vessel safety multi-task optimal tracking control system may further include a communication interface 303 and a bus 300. Figure 7 As shown, the processor 301 , the memory 302 , and the communication interface 303 are connected via a bus 300 and communicate with each other.

[0117] The communication interface 303 is used to implement communication between the various modules, devices, units, and / or devices in the embodiments of the present application. The communication interface 303 can also implement data communication with other components such as: external devices, image / data acquisition equipment, databases, external storage, and image / data processing workstations.

[0118] Bus 300 , which includes hardware, software, or both, couples the components of the inland vessel safety multi-task optimal tracking control system. Bus 300 includes, but is not limited to, at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 300 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Bus 300 may include one or more buses, where appropriate. Although embodiments herein describe and illustrate a particular bus, this application contemplates any suitable bus or interconnect.

[0119] The computer device can execute the inland waterway vessel safety multi-task optimal tracking control method in the embodiment of the present application, thereby realizing the combination of Figure 1 and Figure 2 A safe multi-task optimal tracking control method for inland waterway vessels is described.

[0120] In addition, in conjunction with the inland waterway vessel safety multi-task optimal tracking control method in the above-mentioned embodiments, the present application can provide a computer-readable storage medium for implementation. The computer-readable storage medium stores computer program instructions; when the computer program instructions are executed by a processor, any of the inland waterway vessel safety multi-task optimal tracking control methods in the above-mentioned embodiments is implemented.

[0121] It should be noted that the various technical features of the above-described embodiments can be combined in any manner. To simplify the description, not all possible combinations of the various technical features in the above-described embodiments are described. However, as long as there are no contradictions in the combination of these technical features, they should be considered to be within the scope of this specification. In addition, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0122] Those skilled in the art will readily understand that the above-described embodiments merely represent several implementation methods of the present application, and their descriptions are relatively specific and detailed, but should not be construed as limiting the scope of the invention. It should be noted that a person of ordinary skill in the art may make several variations and improvements without departing from the concept of the present application, and these variations and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be based on the appended claims.

Claims

1. A safe multi-task optimal tracking control method for inland waterway vessels, characterized in that: The method includes: Establish discrete kinematic and dynamic equations for inland waterway vessels and decouple the nonlinear system of the vessel into kinematic and dynamic subsystems; In the kinematic subsystem, an optimal virtual control law is designed to meet the kinematic safety multi-task requirements, including: Combined with the preset trajectory, the kinematic level trajectory tracking error is generated; Define a navigation area set and design a kinematic level ship navigation obstacle function based on the navigation area set; Based on the kinematic-level trajectory tracking error, the control commands of the kinematic subsystem, and the kinematic-level ship navigation obstacle function, a kinematic multi-task objective cost function with safety dynamic constraints is designed. The Hamilton-Jacobi-Bellman equation is constructed, and combined with the back-recursion technique and adaptive dynamic programming algorithm, an ideal virtual control law that meets the optimality requirements is solved. Design a kinematics-level action-evaluation multi-layer neural network architecture to approximate the kinematic optimal multi-task objective cost function and the unknown uncertainties of the ideal virtual control law, obtain the optimal virtual control law, and update the kinematics-level action-evaluation multi-layer neural network weights; Based on the optimal virtual control law, an optimal control law that meets the multi-task requirements of dynamic safety is designed in the dynamic subsystem to guide inland waterway vessels to follow the preset trajectory. Specifically, the following are the steps: Calculate the speed tracking error at the dynamic level based on the optimal virtual control law and ship speed information; Define a set of navigation speed regions and design a dynamic level ship speed obstacle function based on the set of navigation speed regions; Based on the dynamic-level speed tracking error, the control commands of the dynamic subsystem, and the dynamic-level ship speed obstacle function, a dynamic multi-task objective cost function with dynamic speed safety constraints is designed. The Hamilton-Jacobi-Bellman equation is constructed and solved using an adaptive dynamic programming algorithm to obtain the ideal control law that meets the optimality requirements. A dynamic hierarchical action-evaluation multi-layer neural network architecture is designed to approximate the dynamic optimal multi-task objective cost function and the unknown uncertainties of the ideal control law, obtain the optimal control law, guide the inland vessel to track the preset trajectory, and update the dynamic hierarchical action-evaluation multi-layer neural network weights.

2. The inland waterway vessel safety multi-task optimal tracking control method according to claim 1 is characterized in that: The navigation area set includes an absolutely safe navigation area, a relatively safe navigation area and a dangerous navigation area.

3. The inland waterway vessel safety multi-task optimal tracking control method according to claim 2, characterized in that: The design of the kinematic level ship navigation obstacle function based on the navigation area set includes: Determine the navigation area set to which the kinematic level trajectory tracking error belongs: When the kinematic level trajectory tracking error is in the absolutely safe navigation area, a logarithmic obstacle function is designed; When the kinematic-level trajectory tracking error is in a relatively safe navigation area, a composite obstacle function is constructed, the composite obstacle function including a danger repelling function and a safety attracting function, to guide the kinematic-level trajectory tracking error toward the origin and away from the navigation danger boundary; wherein the navigation danger boundary is the boundary between the relatively safe navigation area and the dangerous navigation area; When the kinematic level trajectory tracking error is located in a dangerous navigation area, the barrier function value tends to infinity.

4. The inland waterway vessel safety multi-task optimal tracking control method according to claim 2 or 3, characterized in that: The kinematic multi-task objective cost function with safety dynamic constraints includes a trajectory tracking error penalty function based on the kinematic level trajectory tracking error, a control energy consumption function based on the control command of the kinematic subsystem, and a safety cost function based on the kinematic level ship navigation obstacle function. At the same time, the kinematic level trajectory tracking error satisfies the inland waterway ship safety dynamic constraints, that is, the kinematic level trajectory tracking error is located in a navigation safety set, and the navigation safety set includes an absolutely safe navigation area and a relatively safe navigation area. The updating of kinematic level actions - evaluating multi-layer neural network weights, includes: The normalized gradient descent method is used to derive the weight update law of the kinematic-level action multi-layer neural network and the weight update law of the kinematic-level evaluation multi-layer neural network. The lifelong learning method and the online elastic weight consolidation method are adopted in the weight update law of the kinematic-level action multi-layer neural network and the weight update law of the kinematic-level evaluation multi-layer neural network. The regularization term is introduced and combined with the information diagonal matrix of historical tasks to suppress the catastrophic forgetting of the control strategy in multi-task scenarios. At the same time, the kinematic control obstacle function is integrated as a safety constraint condition in the kinematic-level action multi-layer neural network update law, which cooperates with the kinematic-level ship navigation obstacle function to keep the ship trajectory within the navigation safety set.

5. The inland waterway vessel safety multi-task optimal tracking control method according to claim 1, characterized in that: The navigation speed area set includes an absolutely safe navigation speed area, a relatively safe navigation speed area and a dangerous navigation speed area.

6. The inland waterway vessel safety multi-task optimal tracking control method according to claim 5, characterized in that: The design of the dynamic level ship speed obstacle function based on the navigation speed region set includes: Determine the navigation speed region set to which the dynamic level speed tracking error belongs: When the dynamic level speed tracking error is in the absolutely safe area of ​​navigation speed, a logarithmic barrier function is designed; When the dynamic level speed tracking error is in a relatively safe navigation speed region, a composite obstacle function is constructed, the composite obstacle function including a danger repelling function and a safety attracting function, guiding the dynamic level speed tracking error toward the origin and away from the navigation speed danger boundary; wherein the navigation speed danger boundary is the boundary between the relatively safe navigation speed region and the navigation speed danger region; When the dynamic level speed tracking error is in the navigation speed danger zone, the barrier function value tends to infinity.

7. The inland waterway vessel safety multi-task optimal tracking control method according to claim 5 or 6, characterized in that: The dynamic multi-task objective cost function with dynamic speed safety constraints includes a speed tracking error penalty function of a dynamic level speed tracking error, a control energy consumption function based on a control command of a dynamic subsystem, and a safety cost function based on a dynamic level ship speed obstacle function. At the same time, the dynamic level speed tracking error satisfies the dynamic safety constraints of inland waterway ships, that is, the dynamic level speed tracking error is located in a navigation speed safety set, and the navigation speed safety set includes an absolute navigation speed safety area and a relatively navigation speed safety area. The updating dynamics level action - evaluating the weights of the multi-layer neural network, includes: Using the normalized gradient descent method, the weight update law of the dynamic-level action multi-layer neural network and the weight update law of the dynamic-level evaluation multi-layer neural network are derived. The lifelong learning method and the online elastic weight consolidation method are adopted in the weight update law of the dynamic-level action multi-layer neural network and the weight update law of the dynamic-level evaluation multi-layer neural network. The regularization term is introduced and combined with the information diagonal matrix of the historical tasks to suppress the catastrophic forgetting of the control strategy in the multi-task scenario; at the same time, the dynamic control obstacle function is integrated as a safety constraint condition in the dynamic-level action multi-layer neural network update law, and cooperates with the dynamic-level ship speed obstacle function to keep the ship speed within the safe set of navigation speeds.

8. A multi-task optimal tracking control system for inland waterway vessel safety, characterized in that: The system applies the inland waterway vessel safety multi-task optimal tracking control method according to any one of claims 1 to 7, comprising: Model building and decoupling module, used to establish discrete kinematic and dynamic equations for inland waterway vessels, and decouple the nonlinear system of the vessel into kinematic and dynamic subsystems; The kinematic subsystem control module, including an online learning unit and a safety decision-making unit, generates the optimal virtual control law by minimizing the kinematic multi-task objective cost function with safety dynamic constraints; The dynamics subsystem control module, including an online learning unit and a safety decision-making unit, generates an optimal control law by minimizing the dynamics multi-task objective cost function with speed safety dynamic constraints, and guides the inland waterway vessel to follow the preset trajectory; The online learning unit uses a corresponding level of action-evaluation multi-layer neural network architecture, lifelong learning methods, and elastic weight consolidation algorithms to update network weights in real time. The safety decision-making unit monitors the tracking error of the corresponding level and the distance to the danger zone in real time, and updates the obstacle function of the corresponding level. The danger zone is reflected as the position of the obstacle at the kinematic level, that is, the area that the ship cannot reach, and is reflected as the limit of the ship's navigation speed at the dynamic level.

9. A computer device, characterized in that: include: A processor and a memory, the memory storing programs or instructions that can be run on the processor, and the programs or instructions, when executed by the processor, implementing the steps of the inland vessel safety multi-task optimal tracking control method described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that Programs or instructions are stored thereon, and when the programs or instructions are executed by the processor, the steps of the inland waterway vessel safety multi-task optimal tracking control method described in any one of claims 1 to 7 are implemented.

Images