Air cushion vehicle control method, device, equipment and storage medium

Abstract

The present disclosure provides a hovercraft control method, device, equipment and storage medium, the method comprises: obtaining state data and expected trajectory of the hovercraft, determining the expected motion data of the hovercraft based on the state data and the expected trajectory; determining the sliding mode surface of the hovercraft according to the expected motion data and the state data, determining the basic sliding mode control item based on the sliding mode surface; constructing an input vector based on the sliding mode surface and the state data, estimating the input vector using a first neural network, and obtaining a disturbance compensation item output by the first neural network; superimposing the disturbance compensation item and the basic sliding mode control item to determine the control input for controlling the hovercraft.

Description

Technical Field

[0001] This disclosure relates to the field of intelligent control of hovercraft, and in particular to a hovercraft control method, apparatus, equipment and storage medium. Background Technology

[0002] In actual operation, hovercraft are often in complex sea states and under strong environmental disturbances, and their dynamic systems exhibit characteristics such as high nonlinearity, time-varying parameters, and significant model uncertainties. Affected by external disturbances such as wind, waves, and currents, as well as internal factors such as air cushion leakage and changes in propulsion efficiency, the attitude control accuracy and trajectory tracking performance of hovercraft are prone to decline, and even obvious control oscillations and buffeting phenomena may occur.

[0003] To address the aforementioned issues, this invention proposes a hovercraft control method that integrates reinforcement learning and sliding mode control, thereby improving the hovercraft's control performance and environmental adaptability in complex sea conditions. Summary of the Invention

[0004] This disclosure provides a hovercraft control method, apparatus, device, and storage medium to at least solve the above-mentioned technical problems existing in the prior art.

[0005] According to a first aspect of this disclosure, a hovercraft control method is provided, the method comprising: Acquire the state data and desired trajectory of the hovercraft, and determine the desired motion data of the hovercraft based on the state data and the desired trajectory; The sliding surface of the hovercraft is determined based on the desired motion data and the state data, and basic sliding control terms are determined based on the sliding surface. An input vector is constructed based on the sliding surface and the state data. The input vector is estimated using a first neural network, and the disturbance compensation term output by the first neural network is obtained. The disturbance compensation term and the basic sliding mode control term are superimposed to determine the control input used to control the hovercraft.

[0006] In one possible implementation, determining the desired motion data of the hovercraft based on the state data and the desired trajectory includes: Based on the desired trajectory, determine the desired location information corresponding to the current location information in the state data; The position tracking error is determined based on the current position information, the desired position information, and the yaw angle in the status data; The desired motion data is determined based on the position tracking error, and the desired motion data includes: desired forward velocity and desired yaw angle.

[0007] In one possible implementation, determining the sliding surface of the hovercraft based on the desired motion data and the state data, and determining basic sliding control terms based on the sliding surface, includes: The yaw angle tracking error is determined based on the desired yaw angle and the yaw angle, and the desired angular velocity is determined based on the yaw angle tracking error; The yaw rate tracking error is determined based on the angular velocity of the hovercraft in the state data and the desired angular velocity. The forward velocity tracking error is determined based on the expected forward velocity and the forward velocity in the state data; A first sliding surface is constructed based on the yaw rate tracking error, and a second sliding surface is constructed based on the forward velocity tracking error. A first basic sliding mode control term is determined based on the first sliding mode surface, and a second basic sliding mode control term is determined based on the second sliding mode surface. The first basic sliding mode control term is used to control the eccentric torque of the hovercraft, and the second basic sliding mode control term is used to control the propulsion of the hovercraft.

[0008] In one possible implementation, determining a first basic sliding control term based on the first sliding surface and determining a second basic sliding control term based on the second sliding surface includes: The first sliding mode variable change rate is determined based on the first sliding mode surface; Based on the sliding surface approach condition, determine the first equivalent control term corresponding to the first sliding variable change rate; based on the sliding mode robust control requirements, determine the first switching control term corresponding to the first sliding variable change rate. The first equivalent control term and the first switching control term are superimposed to obtain the first basic sliding mode control term; The rate of change of the second sliding mode variable is determined based on the second sliding surface; Based on the sliding surface approach condition, determine the second equivalent control term corresponding to the second sliding variable change rate; based on the sliding mode robust control requirements, determine the second switching control term corresponding to the second sliding variable change rate. The second equivalent control term and the second switching control term are superimposed to obtain the second basic sliding mode control term.

[0009] In one possible implementation, the step of constructing an input vector based on the sliding surface and the state data, estimating the input vector using a first neural network, and obtaining the disturbance compensation term output by the first neural network includes: Determine the sliding mode variable corresponding to the sliding mode surface, and determine the rate of change of the sliding mode variable based on the sliding mode variable; The state data, the sliding mode variable, and the rate of change of the sliding mode variable are stacked to obtain the input vector; Perturbation pseudo-labels are constructed by observing or estimating the state of the hovercraft, and the first neural network is trained based on the pseudo-labels. The input vector is input into the trained first neural network to obtain the perturbation estimate output by the trained first neural network through forward propagation; The disturbance estimate is mapped according to the preset input channel mapping rules to obtain the disturbance compensation term.

[0010] In one possible implementation, after determining the control input for controlling the hovercraft, the method further includes: A state space vector for reinforcement learning is constructed based on the sliding mode variable, the position tracking error, and the control input. The state space vector is input into the reinforcement learning network to obtain the sliding mode control update parameters and the first neural network update parameters output by the reinforcement learning network from the preset action space through forward propagation. The sliding surface is updated based on the sliding mode control update parameters, and the first neural network is updated based on the first neural network update parameters.

[0011] In one possible implementation, the method further includes: Obtain the state data for the next moment, and determine the immediate reward based on the state data for the next moment and the state data itself; The instant reward, the status data, and the status data at the next moment are stored as a sample data in the experience pool; The reinforcement learning network is updated based on at least one sample data from the experience pool.

[0012] According to a second aspect of this disclosure, a hovercraft control device is provided, the device comprising: A trajectory tracking unit is used to acquire the state data and desired trajectory of the hovercraft, and to determine the desired motion data of the hovercraft based on the state data and the desired trajectory. A sliding mode control unit is used to determine the sliding mode surface of the hovercraft based on the desired motion data and the state data, and to determine basic sliding mode control items based on the sliding mode surface. The disturbance compensation unit is used to construct an input vector based on the sliding surface and the state data, estimate the input vector using a first neural network, and obtain the disturbance compensation term output by the first neural network. The control output unit is used to superimpose the disturbance compensation term and the basic sliding mode control term to determine the control input for controlling the hovercraft.

[0013] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods described in this disclosure.

[0014] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing the computer to perform the methods described in this disclosure.

[0015] This disclosure discloses a hovercraft control method, apparatus, device, and storage medium. By acquiring the hovercraft's state data and desired trajectory, desired motion data is determined. A sliding surface is constructed using the desired motion data to derive basic sliding mode control terms. A disturbance compensation term is constructed using a first neural network. The disturbance compensation term is superimposed with the basic sliding mode control term to obtain the hovercraft's control input. Reinforcement learning is used to update the sliding surface and the first neural network, achieving coordinated control of the hovercraft's yaw angle and speed. The beneficial effects of this method include: effectively reducing trajectory tracking errors, suppressing control oscillations and chattering, and improving the hovercraft's control performance and environmental adaptability in complex sea states and highly uncertain environments, while ensuring system stability and robustness.

[0016] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0017] The above and other objects, features, and advantages of this disclosure will become readily apparent from the following detailed description of exemplary embodiments, taken in conjunction with the accompanying drawings. Several embodiments of this disclosure are illustrated in the drawings by way of example and not limitation, in which: In the accompanying drawings, the same or corresponding reference numerals indicate the same or corresponding parts.

[0018] Figure 1 This illustration shows a schematic flow diagram of the implementation of a hovercraft control method according to an embodiment of the present disclosure. Figure 1 ; Figure 2 This illustration shows a schematic flow diagram of the implementation of a hovercraft control method according to an embodiment of the present disclosure. Figure 2 ; Figure 3 A schematic diagram illustrating the trajectory tracking guidance law principle of an embodiment of this disclosure is shown; Figure 4 A structural diagram of the artificial intelligence compensation item according to an embodiment of this disclosure is shown; Figure 5 A structural diagram of the reinforcement learning section according to an embodiment of this disclosure is shown; Figure 6 The diagram shows a control structure for a hovercraft based on a combination of reinforcement learning and artificial intelligence with robust adaptive sliding mode control, according to an embodiment of this disclosure. Figure 7 A schematic diagram of a hovercraft control device according to an embodiment of the present disclosure is shown; Figure 8 A schematic diagram of the composition structure of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0019] To make the objectives, features, and advantages of this disclosure more apparent and understandable, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort are within the scope of protection of this disclosure.

[0020] Figure 1 This illustration shows a schematic flow diagram of the implementation of a hovercraft control method according to an embodiment of the present disclosure. Figure 1 ,like Figure 1 As shown, the implementation process of a hovercraft control method according to an embodiment of this disclosure includes the following steps: Step 101: Obtain the state data and desired trajectory of the hovercraft, and determine the desired motion data of the hovercraft based on the state data and the desired trajectory.

[0021] In this embodiment, the hovercraft's state data includes: its position, forward velocity, lateral velocity, yaw angle, and yaw rate in the northeast coordinate system. The desired trajectory is the hovercraft's reference position trajectory in the planar coordinate system, with time as the independent variable, including the desired longitudinal and lateral positions at each moment. Based on the desired trajectory, the desired position information corresponding to the current position information in the state data is determined. The position tracking error is determined based on the current position information, the desired position information, and the yaw angle in the state data. Specifically, a position deviation vector in the geodetic coordinate system is calculated based on the current position information and the desired position information. A rotation matrix for coordinate transformation is constructed based on the yaw angle and transposed to obtain a transformation relation matrix. The position deviation vector and the coordinate transformation relation matrix are multiplied to obtain the tracking error. The desired motion data is determined based on the position tracking error, which includes: the desired forward velocity and the desired yaw angle. Specifically, the desired forward velocity and the desired yaw angle can be obtained by taking the derivative of the tracking error with respect to time and combining it with the hovercraft's roll angle data.

[0022] Step 102: Determine the sliding surface of the hovercraft based on the desired motion data and the state data, and determine the basic sliding control terms based on the sliding surface.

[0023] In this embodiment, the yaw angle tracking error is determined based on the desired yaw angle and the yaw angle itself. The desired angular velocity is determined based on the yaw angle tracking error. The yaw angle velocity tracking error is determined based on the angular velocity of the hovercraft in the state data and the desired angular velocity. The dynamic representation of the heading tracking error is obtained by differentiating the yaw angle tracking error. Simultaneously, a virtual control law for the desired yaw angle velocity is designed. Combining the virtual control law and the dynamic representation of the heading tracking error yields the yaw angle velocity tracking error. The forward velocity tracking error is determined based on the desired forward velocity and the forward velocity in the state data. A first sliding surface is constructed based on the yaw angle velocity tracking error, and a second sliding surface is constructed based on the forward velocity tracking error. The first sliding surface represents the integral sliding surface designed for the yaw angle, and the second sliding surface represents the integral sliding surface designed for the speed.

[0024] In this embodiment, a first basic sliding mode control term is determined based on the first sliding mode surface, and a second basic sliding mode control term is determined based on the second sliding mode surface. The first basic sliding mode control term is used to control the eccentric torque of the hovercraft, and the second basic sliding mode control term is used to control the propulsion force of the hovercraft. Specifically, a first sliding mode variable change rate is determined by differentiating the first sliding mode surface over time; a first equivalent control term corresponding to the first sliding mode variable change rate is determined according to the sliding mode surface convergence condition; and a first switching control term corresponding to the first sliding mode variable change rate is determined according to the sliding mode robust control requirement. The sliding mode surface convergence condition is that time approaches 0, and the robust control requirement is that the sliding mode variable and the sliding mode variable change rate ensure that the state converges to the sliding mode surface within a finite time, such that the amplitude of the switching control term covers the upper bound of the disturbance. The first equivalent control term and the first switching control term are superimposed to obtain the first basic sliding mode control term; the second sliding mode variable change rate is determined based on the second sliding surface; the second equivalent control term corresponding to the second sliding mode variable change rate is determined according to the sliding surface convergence condition; the second switching control term corresponding to the second sliding mode variable change rate is determined according to the sliding mode robust control requirements; the second equivalent control term and the second switching control term are superimposed to obtain the second basic sliding mode control term.

[0025] Step 103: Construct an input vector based on the sliding surface and the state data, estimate the input vector using a first neural network, and obtain the disturbance compensation term output by the first neural network.

[0026] In this embodiment, the sliding mode variable corresponding to the sliding surface is determined, and the rate of change of the sliding mode variable is determined based on the sliding mode variable. Specifically, the sliding mode variable (first sliding mode variable, second sliding mode variable) can be obtained by real-time calculation of the sliding surface (first sliding mode surface, second sliding mode surface). The rate of change of the sliding mode variable is obtained by differentiating the sliding mode variable with respect to time. The rate of change of the sliding mode variable includes the aforementioned rate of change of the first sliding mode variable and the rate of change of the second sliding mode variable. The state data, the sliding mode variable, and the rate of change of the sliding mode variable are stacked to obtain the input vector. A perturbation pseudo-label is constructed by observing or estimating the state of the hovercraft. The first neural network is trained based on the pseudo-label. The derivative of the state data with respect to time can be estimated by difference or filtering. The perturbation pseudo-label is constructed by the derivative of the state data with respect to time. The first neural network is a perturbation observation neural network or a perturbation learning neural network. The input vector is input into the trained first neural network to obtain the perturbation estimate output by the trained first neural network through forward propagation. The perturbation estimate is mapped according to a preset input channel mapping rule to obtain the perturbation compensation term.

[0027] Step 104: The disturbance compensation term and the basic sliding mode control term are superimposed to determine the control input used to control the hovercraft.

[0028] In this embodiment, reinforcement learning is used to adaptively adjust the key parameters of sliding mode control and the first neural network. Specifically, after determining the control input for controlling the hovercraft, a state space vector for reinforcement learning is constructed based on the sliding mode variable, the position tracking error, and the control input. The state space vector includes: tracking error, tracking error rate of change, sliding mode variable, rate of change of the sliding mode variable, and rate of change of the control input. The state space vector is input into the reinforcement learning network to obtain the sliding mode control update parameters and the first neural network update parameters output by the reinforcement learning network from a preset action space through forward propagation. The preset action space is a vector defined by the sliding mode switching gain, sliding surface parameters, and weight coefficients of the disturbance compensation term. The sliding surface is updated based on the sliding mode control update parameters, and the first neural network is updated based on the first neural network update parameters. The sliding mode control update parameters correspond to the aforementioned sliding mode switching gain and sliding surface parameters, and the first neural network update parameters correspond to the weight coefficients of the disturbance compensation term.

[0029] In this embodiment, after executing the control input, the state data at the next moment is acquired. Based on the state data at the next moment and the current state data, an immediate reward is determined. Specifically, the tracking error magnitude and control oscillation degree are determined based on the state data at two different moments, and the immediate reward is further calculated. The immediate reward, the current state data, and the state data at the next moment are stored as a sample data in an experience pool. The reinforcement learning network is updated based on at least one sample data in the experience pool. When the maximum number of control steps in a single round or a preset training condition is reached, the current round of training ends, and the above training process is repeated until the preset maximum number of training rounds is reached. The optimized sliding mode control parameters are then output.

[0030] Figure 2 An embodiment of the present disclosure is shown of a hovercraft. Implementation flow diagram of the control method Figure 2 ,like Figure 2 As shown, the implementation process of a hovercraft control method according to an embodiment of this disclosure includes the following steps: Step 201: Establish a nonlinear dynamic model of the hovercraft in the horizontal plane.

[0031] In this embodiment, a nonlinear dynamic model of the hovercraft in the horizontal plane is established. Uncertainties in model parameters, unmodeled hydrodynamics, hovercraft leakage effects, and external wind and wave disturbances are uniformly summarized into the lumped disturbance term of the system. Yaw angle and speed trajectory tracking are clearly defined as the control objectives. Specifically, the kinematic equations are:

[0032] in, and Indicates the position in the northeast coordinate system. Forward velocity, For lateral velocity, Yaw angle Let yaw rate be denoted as yaw angle velocity. Furthermore, uncertainties such as model parameter uncertainties, unmodeled hydrodynamic effects, thrust and damping variations caused by air cushion leakage, and external wind and wave disturbances are uniformly summarized into a lumped disturbance term for the system, simplifying the model structure and enhancing its universality. Based on this, the control objective is defined as achieving high-precision tracking of the hovercraft's yaw angle and speed to a given reference trajectory under the condition of lumped disturbances, providing a dynamic basis for the subsequent design of the controller and reinforcement learning algorithm.

[0033] Step 202: Construct the trajectory tracking guidance law.

[0034] In this embodiment, the position tracking error of the hovercraft is calculated based on the desired trajectory information, and the desired speed and desired yaw angle are generated by the guidance law as reference inputs to the control system. A schematic diagram of the trajectory tracking guidance law principle is shown below. Figure 3 As shown, where, and The coordinate axes are in the northeast coordinate system. and The coordinate axes of the local coordinate system, ( () represents the desired location coordinates. Yaw angle For the desired yaw angle, Let V be the sideslip angle of the hovercraft, and V be the resultant velocity vector. The look-ahead distance represents the distance the current reference point extends forward along the trajectory, used to determine the future reference state.

[0035] Specifically, the position tracking error of a hovercraft is defined as:

[0036] in, and This represents the horizontal and vertical position tracking errors in the local coordinate system. This represents the yaw angle in the reference coordinate system. and Indicates the actual position coordinates. and Represents the expected position coordinates at time t, reference trajectory .

[0037] Position tracking error and Taking the derivative with respect to time yields the following result:

[0038] here, , This is called the sideslip angle of a hovercraft.

[0039] To ensure that the position tracking error eventually converges, the following guidance law was designed:

[0040] here, For the desired speed, For the desired yaw angle, These are the design parameters for the guidance law.

[0041] Step 203: Construct the basic sliding mode control law.

[0042] In this embodiment of the disclosure, the tracking errors and their rates of change of the speed and yaw channels are constructed according to the desired speed and desired yaw angle, respectively. Based on this, a sliding mode surface is designed to complete the initialization of the robust adaptive sliding mode controller. Furthermore, an equivalent control term and a switching control term are constructed according to the sliding mode surface, and an adaptive law is designed to estimate the upper bound of the lumped disturbance online, forming a basic sliding mode control law without intelligent compensation term.

[0043] Specifically, first, we define the yaw angle tracking error:

[0044] Differentiating the above equation, we have:

[0045] Designing virtual control laws:

[0046] in, These are the design parameters in the virtual control law.

[0047] Define the yaw rate tracking error as:

[0048] in This represents the actual yaw rate. To express the desired yaw rate, further design the sliding surface:

[0049] in, Represented as the designed integral sliding surface, Here, S represents the design parameters of the integral sliding surface, and S represents the sliding variable.

[0050] Taking the time derivative of the sliding mode face, we get:

[0051] in, This represents the moment of inertia of the hovercraft about its central axis. It is a lumped disturbance that includes uncertainties in the hovercraft model and environmental disturbances. To overcome its impact on the control system, the yaw moment acting on the hovercraft is... The control law is designed as follows:

[0052] in, For the controller's design parameters, Indicates concentrated interference The upper realm, Indicates concentrated disturbance interference The upper bound of the estimate is obtained using an adaptive estimation method. Therefore, the adaptive law is designed as follows:

[0053] in, These are all design parameters for the control system.

[0054] Similarly, the speed error is defined as:

[0055] in, Indicates longitudinal velocity. This represents the desired longitudinal velocity.

[0056] Design the sliding surface:

[0057] in, The integral sliding surface is designed for the speed of the ship. These are the design parameters for the corresponding integral sliding surface.

[0058] Differentiating the design of the sliding surface with respect to time yields:

[0059] Where m represents the mass of the hovercraft. It is a lumped disturbance that includes uncertainties in the special-purpose ship model and environmental interference. To overcome its impact on the control system, the propulsion force acting on the special-purpose ship... The control law is designed as follows:

[0060] in Design parameters for the controller. Indicates concentrated interference The upper realm, Indicates concentrated disturbance interference The upper bound of the estimate is obtained using an adaptive estimation method. Therefore, the following adaptive law was designed:

[0061] in, These are all design parameters for the control system. Indicates adaptive gain. is the coefficient of the robustness correction term.

[0062] By incorporating the yaw rate tracking error into the sliding surface for real-time calculation, the sliding mode variable can be obtained. The rate of change of the sliding mode variable can be obtained by differentiating it with respect to time. This process is further refined by differentiating the sliding surface with time and ensuring… The equivalent control term can be obtained, and the time derivative of the sliding surface is obtained while ensuring... This allows us to select the switching control item, among which... For sliding mode variables, is the rate of change of the sliding mode variable.

[0063] Step 204: Construct an artificial intelligence perturbation learning compensation term.

[0064] In this embodiment of the disclosure, to reduce the conservatism and chattering of sliding mode control under strong uncertainty conditions, an artificial intelligence disturbance learning model is constructed. A disturbance compensation term is introduced into the basic sliding mode control law, and the artificial intelligence disturbance learning compensation term is added to the control input. Figure 4 A structural diagram of the artificial intelligence compensation item according to an embodiment of this disclosure is shown, as follows: Figure 4 As shown, a neural network is selected as the perturbation estimator, with the system state, sliding mode variable, and rate of change of the sliding mode variable as the input vector. Characterizing the current operating state and sliding mode information of the system, we obtain:

[0065] in, For state information, the neural network output is used to estimate unknown perturbations:

[0066] in, The parameter is Neural network model, To handle disturbances that vary with state information x and time t The estimate is given by t, where t is the time variable.

[0067] Time derivative of state information in engineering implementation It can be obtained by difference / filter estimation, so perturbation pseudo-labels can be constructed for training, and the learning objective of the neural network can be set as minimizing the perturbation observation error. The disturbance estimate is obtained. Then, it is mapped as a disturbance compensation term through the input channel. The equivalent control term, sliding mode switching term, and artificial intelligence disturbance compensation term are superimposed to form a complete control input u, which is then applied to the hovercraft actuator. Specifically, the channel here refers to both the neural network output channel and the control input channel corresponding to the hovercraft actuator; the disturbance estimate output by the network... Belonging to the state space dimension and corresponding to the disturbance dimension, it cannot be directly transmitted to the hovercraft hardware. Channel mapping involves conversion of the control input matrix, gain scaling, dimension alignment, and dimension transformation, while simultaneously combining actuator saturation constraints to achieve amplitude limiting. This transforms the estimated disturbance dimension and dimension into a neural network disturbance compensation term with the same dimension and size as the sliding mode control signal. The superposition operation of multiple control terms after mapping involves algebraically superimposing the equivalent control term, sliding mode switching term, and artificial intelligence disturbance compensation term to form a complete control input that acts on the hovercraft actuator. In this superimposed control structure, the neural network compensation term... It will actively counteract the unknown real disturbances of the system. Originally, all disturbances had to be countered by sliding mode switching terms with large amplitudes. Now, after most disturbances are pre-compensated, the amplitude and gain of the sliding mode switching terms can be reduced, the chattering phenomenon caused by high-frequency switching can be weakened, and the system disturbances are partially offset, thereby reducing the dependence of the sliding mode switching terms on large disturbances and improving tracking accuracy.

[0068] In this embodiment, the artificial intelligence compensation term is only used to estimate and offset unknown disturbances. Its output, together with the switching term, acts on the system, but the system stability is guaranteed by the sliding mode control structure. Even when the neural network estimation error is large or there is a short-term mismatch, the sliding mode switching term can still maintain the system's robustness to disturbances and ensure that the error convergence trend is not disrupted. As the neural network learns and gradually converges, the required switching gain will be significantly reduced, thereby reducing control oscillations and chattering, and improving tracking performance in complex environments.

[0069] Step 205: Introduce reinforcement learning to adjust parameters.

[0070] In this embodiment of the disclosure, Figure 5 A structural diagram of the reinforcement learning portion of an embodiment of this disclosure is shown, as follows: Figure 5 As shown, based on the artificial intelligence perturbation learning compensation term, reinforcement learning is introduced to adaptively adjust the key parameters of the sliding mode controller and the perturbation compensation term. The core of this approach lies in inputting the tracking error and sliding mode variables output from the system's state-space model into the iterative evaluation system. Guided by the reward function that minimizes the tracking error and suppresses control oscillations, a policy network is used to dynamically evaluate and adjust the sliding mode control parameters and artificial intelligence compensation weights within the action space, ultimately generating an optimized set of control parameters and compensation weights. Preferably, the adaptive adjustment problem of control parameters is modeled as a Markov Decision Process (MDP). A Markov Decision Process typically consists of a state space S, an action space A, state transition probabilities P, action rewards R, and discount factors. The state space is selected to reflect the system's tracking performance and control stability as measurable parameters. The tracking error of the hovercraft, the sliding mode variable, and the rate of change of the control input are used as the state inputs for reinforcement learning at period 0, such as:

[0071] in, To track errors, For sliding mode variables, This represents the rate of change of the control input, where k indicates the time k.

[0072] A reinforcement learning agent outputs control parameter adjustments based on the current policy using the action space. The action space can be defined as:

[0073] in, For sliding mode switching gain, This represents the change in gain during sliding mode switching. For sliding surface parameters, These are the variations in the sliding surface parameters. For artificial intelligence disturbance compensation items The weighting coefficients, For artificial intelligence disturbance compensation items The changes in the weighting coefficients are used to update the sliding mode controller parameters and the AI ​​disturbance compensation weights. The updated control law is then executed to obtain the system's state information for the next cycle. Figure 5 In this model, period 0, period k-1, and period k represent the relationship between the number of training rounds in the previous, current, and next periods (each period is one training round). The instant reward is calculated based on indicators such as tracking error magnitude and control oscillation degree. The instant reward, current state information, and next-moment state information are used as a sample, and the sample data is stored in the experience pool. Sample data is randomly drawn from the experience pool to train the reinforcement learning network, updating the parameters of the policy network and value network, and simultaneously updating the target network parameters. It is determined whether the maximum number of control steps in a single round or the preset training conditions have been reached; if not, training continues; if reached, the current round of training ends. This process is repeated until the preset maximum number of training rounds is reached, and the sliding mode control parameters optimized by reinforcement learning are output, achieving stable and high-precision trajectory tracking control of the hovercraft under complex sea conditions and high uncertainty.

[0074] In this embodiment, the reward function is used to guide the reinforcement learning agent to improve control performance while ensuring stability. Taking into account factors such as tracking error, suppression variable amplitude, and chattering, the reward function is typically designed as a "penalty-type" function. That is, the error, sliding mode variable, and control rate of change are all written as penalty terms, and the reward is the negative of these terms. Then, terminal rewards and safety penalties are added. The reward function can be expressed as:

[0075]

[0076]

[0077] in, Total round reward The reward function is in the form of a penalty. The meaning of the reward function is that when the speed error is large, the reward decreases, forcing the agent to choose parameters more conducive to speed convergence; when the yaw error is large, the reward decreases, prompting the system to improve heading tracking accuracy; and when the speed is a sliding mode variable... sliding mode variables of yaw rate When the value is too large, the reward decreases, indicating that the system is far from the sliding surface and the parameters need further optimization; when the rate of change of the control input is too large, the reward decreases, thereby suppressing chattering and oscillation. These are the weighting coefficients. As a terminal reward, when the system error enters a preset small neighborhood, an additional positive reward can be given to accelerate training convergence. For safety reasons, This is a penalty coefficient; if the control parameter is too large, the control input exceeds the limit, or the system state exceeds the safety boundary, the penalty will take effect. To control the maximum allowable amplitude of the input, Maximum permissible deviation of yaw angle.

[0078] The reinforcement learning agent employs a policy function to adjust its actions based on the output parameters of the current state. During training, it constructs diverse training scenarios by sweeping through different system parameter perturbations, wind and wave intensity, and operating conditions, enabling the reinforcement learning policy to possess good generalization and adaptive capabilities. Since reinforcement learning only operates at the control parameter layer and does not directly generate control inputs, the sliding mode controller can still maintain system stability due to its robust structure when the reinforcement learning policy has not yet fully converged or when performance degrades under unknown operating conditions. Simultaneously, the weights of the AI ​​perturbation compensation term are adjusted by reinforcement learning, achieving a balance between performance improvement and safety constraints, thereby ensuring the safe and reliable operation of the system in complex environments.

[0079] Figure 6The diagram illustrates a control structure for a hovercraft based on a combination of reinforcement learning and artificial intelligence with robust adaptive sliding mode control, as described in this disclosure embodiment. Figure 6 As shown, the process begins with the processing of tracking trajectory and heading signals by the sliding mode control section on the left. The upper bound of the disturbance is estimated online using the sliding mode surface and adaptive law. The system then coordinates with the reinforcement learning agent above, which includes a policy network, a value network, and an experience pool, to perform parameter tuning and policy learning. The generated control quantity drives the air cushion actuator after passing through the disturbance compensation module. At the same time, the system judges whether convergence or preset training conditions are met based on real-time feedback. If the conditions are not met, the data is sent back to the experience pool for further iteration. If the conditions are met, the final optimized control result is output.

[0080] Figure 7 A schematic diagram of a hovercraft control device according to an embodiment of the present disclosure is shown, such as... Figure 7 As shown, an embodiment of this disclosure provides a hovercraft control device comprising: The trajectory tracking unit 701 is used to acquire the state data and desired trajectory of the hovercraft, and to determine the desired motion data of the hovercraft based on the state data and the desired trajectory.

[0081] The trajectory tracking unit 701 is further configured to: determine the desired position information corresponding to the current position information in the state data based on the desired trajectory; determine the position tracking error based on the current position information, the desired position information and the yaw angle in the state data; and determine the desired motion data based on the position tracking error, wherein the desired motion data includes: desired forward velocity and desired yaw angle.

[0082] The sliding mode control unit 702 is used to determine the sliding surface of the hovercraft based on the desired motion data and the state data, and to determine basic sliding mode control items based on the sliding surface.

[0083] The sliding mode control unit 702 is further configured to: determine a yaw angle tracking error based on the desired yaw angle and the yaw angle; determine a desired angular velocity based on the yaw angle tracking error; determine a yaw angle velocity tracking error based on the angular velocity of the hovercraft in the state data and the desired angular velocity; determine a forward velocity tracking error based on the desired forward velocity and the forward velocity in the state data; construct a first sliding mode surface based on the yaw angle velocity tracking error; construct a second sliding mode surface based on the forward velocity tracking error; determine a first basic sliding mode control term based on the first sliding mode surface; and determine a second basic sliding mode control term based on the second sliding mode surface. The first basic sliding mode control term is used to control the eccentric torque of the hovercraft, and the second basic sliding mode control term is used to control the propulsion force of the hovercraft.

[0084] The sliding mode control unit 702 is further configured to: determine a first sliding mode variable change rate based on the first sliding surface; determine a first equivalent control term corresponding to the first sliding mode variable change rate according to the sliding surface convergence condition; determine a first switching control term corresponding to the first sliding mode variable change rate according to the sliding mode robust control requirement; superimpose the first equivalent control term and the first switching control term to obtain the first basic sliding mode control term; determine a second sliding mode variable change rate based on the second sliding surface; determine a second equivalent control term corresponding to the second sliding mode variable change rate according to the sliding surface convergence condition; determine a second switching control term corresponding to the second sliding mode variable change rate according to the sliding mode robust control requirement; and superimpose the second equivalent control term and the second switching control term to obtain the second basic sliding mode control term.

[0085] The disturbance compensation unit 703 is used to construct an input vector based on the sliding surface and the state data, estimate the input vector using a first neural network, and obtain the disturbance compensation term output by the first neural network.

[0086] The disturbance compensation unit 703 is further configured to: determine the sliding mode variable corresponding to the sliding mode surface; determine the sliding mode variable change rate based on the sliding mode variable; stack the state data, the sliding mode variable, and the sliding mode variable change rate to obtain the input vector; construct disturbance pseudo-labels by performing state observation or state estimation on the hovercraft; train the first neural network based on the pseudo-labels; input the input vector into the trained first neural network to obtain the disturbance estimate value output by the trained first neural network through forward propagation; and map the disturbance estimate value according to a preset input channel mapping rule to obtain the disturbance compensation term.

[0087] The control output unit 704 is used to superimpose the disturbance compensation term and the basic sliding mode control term to determine the control input for controlling the hovercraft.

[0088] The reinforcement learning unit 705 is used to construct a state space vector for reinforcement learning based on the sliding mode variable, the position tracking error, and the control input; input the state space vector into the reinforcement learning network to obtain the sliding mode control update parameters and the first neural network update parameters output by the reinforcement learning network from a preset action space through forward propagation; update the sliding surface based on the sliding mode control update parameters, and update the first neural network based on the first neural network update parameters.

[0089] The reinforcement learning unit 705 is further configured to: acquire state data at the next time step; determine an immediate reward based on the state data at the next time step and the state data; store the immediate reward, the state data, and the state data at the next time step as sample data in an experience pool; and update the reinforcement learning network based on at least one sample data in the experience pool.

[0090] In an exemplary embodiment, the trajectory tracking unit 701, sliding mode control unit 702, disturbance compensation unit 703, control output unit 704, and reinforcement learning unit 705 may be implemented by one or more central processing units (CPUs), graphics processing units (GPUs), application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, micro controller units (MCUs), microprocessors, or other electronic components.

[0091] Regarding the apparatus in the above embodiments, the specific manner in which each module and unit performs its operations has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0092] According to embodiments of this disclosure, this disclosure also provides an electronic device and a readable storage medium.

[0093] Figure 8 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0094] like Figure 8As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. The RAM 803 may also store various programs and data required for the operation of the electronic device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0095] Multiple components in electronic device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of displays, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows electronic device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0096] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 performs the various methods and processes described above, such as a hovercraft control method. For example, in some embodiments, a hovercraft control method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of a hovercraft control method described above can be performed. Alternatively, in other embodiments, the computing unit 801 can be configured to perform a hovercraft control method by any other suitable means (e.g., by means of firmware).

[0097] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0098] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0099] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0100] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0101] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0102] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0103] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0104] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0105] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A method for controlling a hovercraft, characterized in that, The method includes: Acquire the state data and desired trajectory of the hovercraft, and determine the desired motion data of the hovercraft based on the state data and the desired trajectory; The sliding surface of the hovercraft is determined based on the desired motion data and the state data, and basic sliding control terms are determined based on the sliding surface. An input vector is constructed based on the sliding surface and the state data. The input vector is estimated using a first neural network, and the disturbance compensation term output by the first neural network is obtained. The disturbance compensation term and the basic sliding mode control term are superimposed to determine the control input used to control the hovercraft.

2. The hovercraft control method according to claim 1, characterized in that, The step of determining the desired motion data of the hovercraft based on the state data and the desired trajectory includes: Based on the desired trajectory, determine the desired location information corresponding to the current location information in the state data; The position tracking error is determined based on the current position information, the desired position information, and the yaw angle in the status data; The desired motion data is determined based on the position tracking error, and the desired motion data includes: desired forward velocity and desired yaw angle.

3. The hovercraft control method according to claim 2, characterized in that, The step of determining the sliding surface of the hovercraft based on the desired motion data and the state data, and determining basic sliding control terms based on the sliding surface, includes: The yaw angle tracking error is determined based on the desired yaw angle and the yaw angle, and the desired angular velocity is determined based on the yaw angle tracking error; The yaw rate tracking error is determined based on the angular velocity of the hovercraft in the state data and the desired angular velocity. The forward velocity tracking error is determined based on the expected forward velocity and the forward velocity in the state data; A first sliding surface is constructed based on the yaw rate tracking error, and a second sliding surface is constructed based on the forward velocity tracking error. A first basic sliding mode control term is determined based on the first sliding mode surface, and a second basic sliding mode control term is determined based on the second sliding mode surface. The first basic sliding mode control term is used to control the eccentric torque of the hovercraft, and the second basic sliding mode control term is used to control the propulsion of the hovercraft.

4. The hovercraft control method according to claim 3, characterized in that, The step of determining the first basic sliding control term based on the first sliding surface and the second basic sliding control term based on the second sliding surface includes: The first sliding mode variable change rate is determined based on the first sliding mode surface; Based on the sliding surface approach condition, determine the first equivalent control term corresponding to the first sliding variable change rate; based on the sliding mode robust control requirements, determine the first switching control term corresponding to the first sliding variable change rate. The first equivalent control term and the first switching control term are superimposed to obtain the first basic sliding mode control term; The rate of change of the second sliding mode variable is determined based on the second sliding surface; Based on the sliding surface approach condition, determine the second equivalent control term corresponding to the second sliding variable change rate; based on the sliding mode robust control requirements, determine the second switching control term corresponding to the second sliding variable change rate. The second equivalent control term and the second switching control term are superimposed to obtain the second basic sliding mode control term.

5. The hovercraft control method according to claim 1, characterized in that, The step of constructing an input vector based on the sliding surface and the state data, estimating the input vector using a first neural network, and obtaining the disturbance compensation term output by the first neural network includes: Determine the sliding mode variable corresponding to the sliding mode surface, and determine the rate of change of the sliding mode variable based on the sliding mode variable; The state data, the sliding mode variable, and the rate of change of the sliding mode variable are stacked to obtain the input vector; Perturbation pseudo-labels are constructed by observing or estimating the state of the hovercraft, and the first neural network is trained based on the pseudo-labels. The input vector is input into the trained first neural network to obtain the perturbation estimate output by the trained first neural network through forward propagation; The disturbance estimate is mapped according to the preset input channel mapping rules to obtain the disturbance compensation term.

6. The hovercraft control method according to claim 2, characterized in that, After determining the control inputs for controlling the hovercraft, the method further includes: A state space vector for reinforcement learning is constructed based on the sliding mode variable, the position tracking error, and the control input. The state space vector is input into the reinforcement learning network to obtain the sliding mode control update parameters and the first neural network update parameters output by the reinforcement learning network from the preset action space through forward propagation. The sliding surface is updated based on the sliding mode control update parameters, and the first neural network is updated based on the first neural network update parameters.

7. The hovercraft control method according to claim 6, characterized in that, The method further includes: Obtain the state data for the next moment, and determine the immediate reward based on the state data for the next moment and the state data itself; The instant reward, the status data, and the status data at the next moment are stored as a sample data in the experience pool; The reinforcement learning network is updated based on at least one sample data from the experience pool.

8. A hovercraft control device, characterized in that, The device includes: A trajectory tracking unit is used to acquire the state data and desired trajectory of the hovercraft, and to determine the desired motion data of the hovercraft based on the state data and the desired trajectory. A sliding mode control unit is used to determine the sliding mode surface of the hovercraft based on the desired motion data and the state data, and to determine basic sliding mode control items based on the sliding mode surface. The disturbance compensation unit is used to construct an input vector based on the sliding surface and the state data, estimate the input vector using a first neural network, and obtain the disturbance compensation term output by the first neural network. The control output unit is used to superimpose the disturbance compensation term and the basic sliding mode control term to determine the control input for controlling the hovercraft.

9. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images