Hovercraft and parameter integrated design method, device and equipment thereof

Abstract

The invention discloses a hovercraft and a parameter integrated design method, device and equipment thereof, and the method comprises the steps: fixing the basic structure parameters of a propeller of the hovercraft, and randomly sampling the strategy hyper-parameters of different controllers on the basis of the basic structure parameters; executing the navigation task in the simulation environment based on different performance costs to obtain corresponding performance costs; a self-adaptive optimization method based on historical gradient estimation is adopted to analyze the variation trend with the system performance, and iterative updating is carried out according to the variation trend until the performance cost numerical value is converged, so that an optimal strategy hyper-parameter is obtained; fixing of the basic structure parameters is relieved, the basic structure parameters and the basic structure parameters jointly form a design vector, the optimal strategy hyper-parameters serve as the benchmark, a two-stage Bayesian optimization strategy is adopted to conduct collaborative search on the design vector, and a global optimal design vector is obtained; and carrying out integrated design of the hovercraft based on the optimal design vector.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of air cushion vehicles, and particularly relates to an air cushion vehicle and a parameter integrated design method, device and equipment thereof. BACKGROUND

[0002] As a unique waterway vehicle, the air cushion vehicle realizes efficient navigation on water or land by air cushion effect. Unlike the single propulsion mode of traditional ships, the air cushion vehicle synchronously completes turning and attitude control through a vector thrust system, which makes the layout parameters of the propeller (including the transverse position, longitudinal position and thrust deflection angle relative to the center of mass of the ship body) the core elements determining the overall performance. The transverse position of the propeller is directly related to the length of the turning force arm, and thus significantly affects the turning response speed and flexibility of the ship; and the combination of the longitudinal position and the thrust deflection angle jointly adjusts the dynamic coupling relationship between the longitudinal propulsion force, the transverse force and the yawing moment, which plays a fundamental role in maintaining straight-horizon stability and resisting wind and wave interference. In practical application scenarios such as unmanned rescue, the air cushion vehicle needs to perform various complex tasks, and its performance is highly dependent on the close cooperation of structural parameters and control strategies.

[0003] However, the air cushion vehicle design process in the prior art generally adopts a mode of separating structural parameter optimization and control strategy design. In the structural design stage, engineers determine the propeller distribution parameters according to experience or simplified models, and then develop a controller based on the fixed structure in the control design stage. This fragmented method fails to fully consider the dynamic influence of structural parameter changes on control performance, for example, when the propeller position is adjusted, the original control strategy may not be able to adapt to the new dynamic characteristics, resulting in problems such as increased trajectory deviation, decreased stability or weakened anti-interference ability in the navigation task. Especially in complex water environments, the strong coupling effect between structure and control parameters is ignored, which makes the air cushion vehicle have obvious deficiencies in straight-horizon stability and wind and wave resistance, and is difficult to meet the needs of high-reliability tasks such as unmanned rescue. SUMMARY

[0004] Therefore, the purpose of the present application is to provide an air cushion vehicle and a parameter integrated design method, device and equipment thereof to improve the above problems.

[0005] The present application provides an air cushion vehicle parameter integrated design method, which comprises: S101, fixing the basic structural parameters of the propeller of the air cushion vehicle, and randomly sampling different strategy hyperparameters of the controller based on the basic structural parameters ; S102, based on different strategy hyperparameters , performing a preset navigation task in a simulation environment to obtain a corresponding performance cost ; S103, using an adaptive optimization method based on historical gradient estimation to analyze the strategy hyperparameters and the change trend of performance cost , and iteratively updating the strategy hyperparameters according to the change trend of performance cost until the performance cost converges, obtaining the optimal strategy hyperparameters ; S104, releasing the fixing of the infrastructure parameters, and combining the infrastructure parameters and the strategy hyperparameters to form a design vector, using the optimal strategy hyperparameters as a reference, using a two-stage Bayesian optimization strategy to cooperatively search the design vector, obtaining the globally optimal design vector; S105, based on the optimal design vector, performing integrated design of the structure and control of the air cushion vehicle.

[0006] Preferably, the infrastructure parameters at least include the propeller angle of the air cushion vehicle propeller; the propeller angle is initially fixed at 0 o .

[0007] Preferably, the navigation task includes straight-line navigation, eight-character navigation, obstacle avoidance navigation, point tracking task, and sine wave course.

[0008] Preferably, the two-stage Bayesian optimization strategy in step S104 specifically includes: first stage fine-tuning: according to the infrastructure parameters and the corresponding strategy hyperparameters to form an initial design vector, using the optimal strategy hyperparameters as a reference, forming a first design space according to a preset search range, and using a Bayesian optimization method based on a Gaussian process surrogate model and an expected improvement acquisition function to perform optimization in the first design space, obtaining a preliminary optimal design vector; second stage fine-tuning: taking the preliminary optimal design vector as the center, narrowing the search range of the infrastructure parameters and the controller hyperparameters, forming a second design space, constructing a plurality of navigation tasks with different difficulty levels for evaluation, and continuing to use the Bayesian optimization method to perform optimization on the average performance of the plurality of navigation tasks with different difficulty levels in the second design space, obtaining the globally optimal design vector.

[0009] Preferably, the specific steps of the Bayesian optimization method include: Initialization: Within the design space, initial sample point set is generated by Latin hypercube sampling, and its performance cost is obtained by simulation evaluation, which constitutes the initial observation data set; Iterative optimization: In each iteration, the Gaussian process model is constructed based on the current observation data set, and the next design vector to be evaluated is recommended by maximizing the expected improvement acquisition function; Evaluation and update: The performance cost of the design vector to be evaluated is obtained by simulation evaluation, and the new data points generated by simulation evaluation are added to the observation data set to update the Gaussian process model; Convergence judgment: When the maximum number of iterations is reached, or the expected improvement value output by the expected improvement acquisition function is lower than the threshold value, or the improvement rate of the performance cost of the continuous iteration is lower than the threshold value, the iteration is terminated, and the design vector with the best performance is selected from the historical data as the output.

[0010] Preferably, the strategy hyperparameter α is a scalar continuously taking values in the interval (0, 1), which continuously adjusts the state error weight matrix and the control input weight matrix of the linear parameter time-varying model predictive controller through a predefined functional relationship, so as to realize the smooth trade-off from the preference trajectory tracking performance to the preference system robustness.

[0011] Preferably, it further comprises: By fusing a data-driven LSTM residual predictor into the LPV-MPC controller, an adaptive controller with online prediction-correction capability is constituted; Based on the adaptive controller, disturbances are added in the simulation environment, and the globally optimal design vector is corrected in the navigation task.

[0012] The embodiment of the application also provides a hovercraft parameter integrated design device, which comprises: A random sampling unit is configured to fix the basic structure parameters of the propeller of the hovercraft, and to randomly sample different strategy hyperparameters of the controller based on the basic structure parameters ; A performance cost calculation unit is configured to execute a preset navigation task in a simulation environment based on the different strategy hyperparameters , and obtain corresponding performance costs ; An adaptive optimization unit is configured to analyze the change trend between the strategy hyperparameters and the system performance by using an adaptive optimization method based on historical gradient estimation, and to iteratively update the strategy hyperparameters until the performance cost converges, so as to obtain optimal strategy hyperparameters ; A collaborative search unit is configured to release the fixing of the basic structure parameters, and to combine the basic structure parameters with the strategy hyperparameters to constitute a design vector, and the optimal strategy hyperparameter As a benchmark, a two-stage Bayesian optimization strategy is used to cooperatively search the design vector to obtain a globally optimal design vector. An integrated design unit is configured to perform integrated design of the structure and control of the air cushion vehicle based on the optimal design vector.

[0013] The embodiment of the present application also provides an air cushion vehicle parameter integrated design device, which comprises a memory and a processor, and the memory stores a computer program which can be executed by the processor to realize the air cushion vehicle parameter integrated design method as described above.

[0014] The embodiment of the present application also provides an air cushion vehicle which is designed and generated by using the air cushion vehicle parameter integrated design method as described above.

[0015] To sum up, the embodiment proposes a collaborative design method based on hierarchical optimization for the optimization problem of strong coupling between the structure and control parameters of the air cushion vehicle. The method breaks the traditional serial design paradigm of "structure first, then control", obtains a robust basic controller through "pre-training", and then cooperatively fine-tunes the structure layout and control strategy in a unified design space through "two-stage Bayesian optimization", thereby realizing deep joint optimization of the two. The embodiment can systematically reveal the inherent trade-off relationship between the structure parameters and the control hyperparameters, thereby significantly improving the trajectory tracking accuracy, maneuverability and overall robustness of the air cushion vehicle in complex rescue missions. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a flowchart of the air cushion vehicle parameter integrated design method provided by the first embodiment of the present application; Figure 2 is a simulation schematic diagram of the air cushion vehicle and its key parameters provided by the first embodiment of the present application; Figure 3 is a principle schematic diagram of the pre-training stage provided by the embodiment of the present application; Fig. 4 is a result schematic diagram of the pre-training stage provided by the embodiment of the present application; wherein Fig. 4(a) is a parameter evolution diagram in the training process; Fig. 4(b) is a diagram showing the change trend of different task costs in the training process; Fig. 4(c) is a diagram showing the relationship between the parameters and the comprehensive cost; and Fig. 4(d) is a diagram showing the cost distribution of the point-to-point task and the sine wave task. Fig. 4(b) is a diagram showing the change trend of different task costs in the training process; Fig. 4(c) is a diagram showing the relationship between the parameters and the comprehensive cost; and Fig. 4(d) is a diagram showing the cost distribution of the point-to-point task and the sine wave task. Figure 5 is a diagram of the target function model of the two-stage Bayesian optimization provided by the embodiment of the present application; Figure 6 is a schematic diagram of the result of a fine-tuning step provided in an embodiment of the present invention; wherein, Figure 6(a) is a diagram of the two-stage Bayesian optimization objective function model during a fine-tuning step; Figure 6(b) is a schematic diagram of the calculation of the fine-tuning function; Figure 6(c) is... Parameter optimization trajectory diagram; Figure 6(d) is a comparison diagram of cost evolution; Figure 7 is a schematic diagram of the results of the two-stage fine-tuning provided in the embodiment of the present invention; wherein, Figure 7(a) is a diagram of the two-stage Bayesian optimization objective function model in the two-stage fine-tuning process; Figure 7(b) is a comparison diagram of the evolution of costs for different tasks; Figure 8 This is a schematic diagram of the integrated design device for hovercraft parameters provided in the second embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 The first embodiment of the present invention provides an integrated design method for hovercraft parameters, which can be executed by an integrated design device for hovercraft parameters (hereinafter referred to as the design device), specifically by one or more processors within the design device, to achieve the following steps: S101, fix the basic structural parameters of the hovercraft's propulsion system, and randomly sample the hyperparameters of different controller strategies based on the basic structural parameters.

[0019] In this embodiment, the integrated design method for hovercraft parameters can be composed of three stages: a pre-training stage, a first-stage fine-tuning stage, and a second-stage fine-tuning stage. The pre-training stage is described in steps S101-S103, while the first and second-stage fine-tuning stages are described in step S104.

[0020] In this embodiment, the hovercraft mainly includes a hull, an air cushion at the bottom of the hull, and a propeller located at the stern of the hull. The propeller is the main power source of the hovercraft.

[0021] The basic structural parameters may include various parameters of the hovercraft, such as hull length, air cushion height at the bottom of the hull, and the number and angle of the thrusters located at the stern. Specifically, in this embodiment, the main basic structural parameter considered is the fan angle of the thrusters, and the fan angle of the thrusters is fixed. It is 0 degrees. Please refer to the following for details.Figure 2 .

[0022] In the embodiment, the strategy hyperparameter is a scalar continuously taking values in the interval (0, 1), which continuously adjusts the state error weight matrix and the control input weight matrix of the linear parameter-varying model predictive controller through a predefined function relationship, to achieve a smooth trade-off from the preference trajectory tracking performance to the preference system robustness.

[0023] In the embodiment, the random sampling can be understood as a process of generating a plurality of candidate values according to a uniform distribution or a normal distribution from a predefined design variable range.

[0024] S102, based on different strategy hyperparameters, performing a predetermined navigation task in a simulation environment to obtain a corresponding performance cost.

[0025] In the embodiment, after fixing the infrastructure parameters, the air cushion vehicle performs a predetermined navigation task in the simulation environment according to the adopted strategy hyperparameter , and a corresponding performance cost .

[0026] The navigation task can include straight-line navigation, eight-character navigation, obstacle avoidance navigation, point tracking task, sine wave course, etc., which are not limited by the present application.

[0027] The performance cost is calculated from the performance parameter value and the corresponding weight parameter. Different navigation tasks can be configured with different performance parameter values and weight parameters. One possible case can refer to Table 1: Table 1 Task weight parameter setting

[0028] S103, using an adaptive optimization method based on historical gradient estimation to analyze the trend of change between the strategy hyperparameter and the system performance, and iteratively updating the strategy hyperparameter accordingly until the performance cost converges, to obtain the optimal strategy hyperparameter.

[0029] In the embodiment, as shown in Figure 3 and FIG. 4, an adaptive optimization method based on historical gradient estimation can be used to analyze the trend of change between the strategy hyperparameter and the performance cost , and iteratively update accordingly until the performance cost (i.e. Figure 3 in J eval (X)) converges. This stage outputs a well-tuned infrastructure parameter C0 ( And find the optimal policy hyperparameters. This completes the pre-training phase. Figure 4(a) shows the training process. The parameter evolution diagram, Figure 4(b) shows the trend of cost changes for different tasks during training, and Figure 4(c) shows... The relationship between parameters and overall cost is shown in Figure 4(d), which illustrates the cost distribution for point-to-point and sinusoidal wave tasks. Figures 4(a) and 4(b) show that in the 10th round, the performance cost... It began to converge in the 19th round. The value gradually stabilizes, eventually converging to 0.738. This reveals that in the traditional configuration without structured assistance, the system employs a somewhat aggressive control strategy to achieve trajectory tracking. The cost distributions in Figures 4(c) and 4(d) also corroborate this, when... At lower levels, the cost of sinusoidal trajectory tracking increases significantly. This stage yields... This represents the performance limitations of the serial design paradigm and serves as a starting point and benchmark for collaborative optimization.

[0030] S104, release the fixation of the basic structure parameters, and combine the basic structure parameters and the strategy hyperparameters to form a design vector. Based on the optimal strategy hyperparameters, use a two-stage Bayesian optimization strategy to perform a collaborative search on the design vector to obtain the globally optimal design vector.

[0031] In this embodiment, as Figure 5 As shown, it is a demonstration diagram of a two-stage Bayesian optimization strategy, using hyperparameters. ,angle and performance cost Given three spatial vector coordinates, print the coordinate points determined based on the training results in three-dimensional space, then plot the two-dimensional surface of the function model using the complex coordinate points, and infer the optimal hyperparameters. ,angle The coordinates of the location. The two-stage Bayesian optimization strategy specifically includes: the first stage fine-tuning (one stage of fine-tuning): constructing an initial design vector based on the basic structural parameters and their corresponding policy hyperparameters, and then optimizing the policy hyperparameters. Based on the first search range, a first design space is formed, and a Bayesian optimization method based on the Gaussian process surrogate model and the desired improved acquisition function is used to find the optimal design vector.

[0032] Specifically, at this stage, the pre-trained optimal policy hyperparameters are used. Based on this, within a large search range, Bayesian optimization is used to quickly locate two parameter regions (hyperparameters) with excellent performance. Fan angle Find the minimum point of the surface graph of performance cost J. The x and y coordinates at the point where performance cost J is optimal are the optimal hyperparameters. and angle The fine-tuning results are shown in Figure 6. In Figure 6, Figure 6(a) shows the two-stage Bayesian optimization objective function model during a fine-tuning process, Figure 6(b) shows a schematic diagram of the fine-tuning function calculation, and Figure 6(c) shows… The parameter optimization trajectory is shown in Figure 6(d), which is a comparison of cost evolution. As can be seen from Figure 6(a), the optimization process efficiently finds a better solution near the pre-training starting point through two-stage Bayesian optimization. The value was then finely adjusted from 0.378 to 0.746. Figure 6(b) shows the decreasing trend of the performance cost function calculated by the function, which approaches convergence after the 14th calculation. Figure 6(c) shows... The parameters begin to converge after 20 rounds and approach stability after 28 rounds. This can be seen from Figure 6(d). The parameters begin to converge after 20 rounds and approach stability after 28 rounds.

[0033] The second stage of fine-tuning (two-stage fine-tuning): Centered on the preliminary optimal design vector, the search range of basic structural parameters and controller hyperparameters is narrowed to form a second design space. Navigation tasks with multiple difficulty levels are constructed for evaluation. Within the second design space, the Bayesian optimization method is used to optimize the average performance of the navigation tasks with multiple difficulty levels to obtain the global optimal design vector.

[0034] Specifically, the two-stage fine-tuning is based on the result of the first-stage fine-tuning (optimal hyperparameters). and angle The search range was narrowed down to the vicinity of [missing information], and tasks with varying difficulties and scenarios were introduced for evaluation. Bayesian optimization was used for fine-tuning to obtain the optimal structure-control parameter combination that performs robustly under different difficulty conditions, i.e., the globally optimal design vector. The fine-tuning results are shown in Figure 7. In Figure 7, Figure 7(a) shows the two-stage Bayesian optimization objective function model during the two-stage fine-tuning process, and Figure 7(b) shows a comparison of the cost evolution for different tasks. As can be seen from Figures 7(a) and 7(b), the performance cost begins to converge and stabilizes in the 12th round, eventually converging stably to [missing information]. =0.766, =6.266°, corresponding to a total cost of 28.946.

[0035] The Bayesian optimization method is required in both the first-stage and second-stage fine-tuning processes. The specific steps of the Bayesian optimization method include: Initialization: Within the design space, an initial sample point set is generated using Latin hypercube sampling, and its performance cost is obtained through simulation evaluation, forming the initial observation dataset; Iterative optimization: In each iteration, a Gaussian process model is constructed based on the current observation dataset. By maximizing the expected improvement acquisition function, the next design vector to be evaluated is recommended; Evaluation and update: The recommended design vector is evaluated through simulation to obtain its performance cost, and the new data points generated by the simulation evaluation are added to the observation dataset to update the Gaussian process model; Convergence judgment: When the maximum number of iterations is reached, or the expected improvement value of the expected improvement acquisition function output is lower than a threshold, or the improvement rate of the performance cost of continuous iterations is lower than a threshold, the iteration terminates, and the design vector with the best performance is selected from the historical data as the output.

[0036] S105, Based on the optimal design vector, perform integrated structural and control design of the hovercraft.

[0037] Based on the above embodiments, it also includes: By integrating a data-driven LSTM residual predictor into the LPV-MPC controller, an adaptive controller with online prediction-correction capability is constructed. Based on the adaptive controller, disturbances are introduced into the simulation environment, and the global optimal design vector is corrected during the navigation mission.

[0038] Specifically, the LPV-MPC controller refers to a model predictive controller based on a linear variable parameter model, which can be implemented using multivariable control algorithms to improve the dynamic response performance and robustness of the system. The LSTM residual predictor can be understood as a time series prediction module based on a long short-term memory network. It can capture dynamic changes in the system caused by external disturbances by learning from historical data, aiming to enhance the controller's adaptability to external disturbances. The adaptive controller refers to a composite control mechanism combining the above two technologies, which can adjust the control output in real time to cope with dynamic disturbances in complex environments.

[0039] In detail, the overall operating logic of this embodiment lies in organically combining the LPV-MPC controller and the LSTM residual predictor to form an adaptive control architecture with online prediction and correction capabilities. In this architecture, the LSTM residual predictor, through learning from historical operating data, can predict the system's dynamic residuals caused by external disturbances such as wind and waves, and feeds the prediction results back to the LPV-MPC controller, thereby dynamically adjusting its control output. The advantage of this design is that the LSTM residual predictor does not rely on a precise physical model, but rather learns disturbance patterns in a data-driven manner, thus maintaining stable control under unknown disturbance conditions.

[0040] Furthermore, disturbances such as wind and waves are actively introduced into the simulation environment. The performance of the adaptive controller is evaluated in real time by executing navigation tasks, and the evaluation results are used to adjust the globally optimal design vector. This process not only verifies the robustness of the design parameters but also integrates the disturbance response into the optimization loop, ensuring that the hovercraft continuously optimizes its straight-line stability and wind and wave resistance in dynamic environments. Based on this, the design of the aforementioned adaptive controller, combined with a Bayesian optimization-based fine-tuning process, further improves the reliability and performance of the hovercraft in complex environments.

[0041] In summary, this embodiment addresses the optimization challenge of the strong coupling between the structure and control parameters of hovercraft by proposing a collaborative design method based on hierarchical optimization. This method breaks away from the traditional sequential design paradigm of "structure first, control later," obtaining a robust basic controller through "pre-training," and then using "two-stage Bayesian optimization" to collaboratively fine-tune the structural layout and control strategy within a unified design space, achieving deep joint optimization of the two. This embodiment can systematically reveal the inherent trade-off between structural parameters and control hyperparameters, thereby significantly improving the trajectory tracking accuracy, maneuverability, and overall robustness of hovercraft in complex rescue missions.

[0042] Please see Figure 8 The second embodiment of the present invention also provides an integrated design device for hovercraft parameters, which includes: The random sampling unit 210 is used to fix the basic structural parameters of the hovercraft's propulsion system and, based on these basic structural parameters, randomly sample the hyperparameters of different controller strategies. ; Performance cost calculation unit 220, used for different strategy hyperparameters The preset navigation task is executed in a simulation environment, and the corresponding performance cost is obtained. ; The adaptive optimization unit 230 is used to analyze the policy hyperparameters using an adaptive optimization method based on historical gradient estimation. The trend of changes in system performance is analyzed, and the strategy hyperparameters are updated iteratively accordingly. Until the performance cost After numerical convergence, the optimal policy hyperparameters are obtained. ; The cooperative search unit 240 is used to unfix the basic structure parameters and combine the basic structure parameters with the strategy hyperparameters. Together they form a design vector, with the optimal strategy hyperparameters Based on this, a two-stage Bayesian optimization strategy is used to collaboratively search the design vector to obtain the globally optimal design vector; The integrated design unit 250 is used for the integrated design of the structure and control of the hovercraft based on the optimal design vector.

[0043] The third embodiment of the present invention also provides an integrated design device for hovercraft parameters, which includes a memory and a processor. The memory stores a computer program, which can be executed by the processor to realize the integrated design method for hovercraft parameters as described above.

[0044] The fourth embodiment of the present invention also provides a hovercraft, which is designed and generated using the hovercraft parameter integrated design method described above.

[0045] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the core principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for integrated parameter design of a hovercraft, characterized in that, include: S101, fix the basic structural parameters of the hovercraft's propulsion system, and based on these basic structural parameters, randomly sample the strategy hyperparameters of different controllers. ; S102, based on different strategy hyperparameters The preset navigation task is executed in a simulation environment, and the corresponding performance cost is obtained. ; S103 employs an adaptive optimization method based on historical gradient estimation to analyze the policy hyperparameters. Performance cost The changing trend between them is analyzed, and the strategy hyperparameters are updated iteratively accordingly. Until the performance cost After numerical convergence, the optimal policy hyperparameters are obtained. ; S104, release the fixation on the basic structure parameters and link the basic structure parameters with the strategy hyperparameters. Together they form a design vector, with the optimal strategy hyperparameters For reference, a two-stage Bayesian optimization strategy is used to perform a collaborative search on the design vector to obtain the globally optimal design vector; S105, Based on the optimal design vector, perform integrated structural and control design of the hovercraft.

2. The integrated design method for hovercraft parameters according to claim 1, characterized in that, The basic structural parameters include at least the propulsion angle of the hovercraft's propulsion system; the propulsion angle is initially fixed at 0. o .

3. The integrated design method for hovercraft parameters according to claim 1, characterized in that, The navigation missions include straight-line navigation, figure-eight navigation, obstacle avoidance navigation, point tracking missions, and sine wave routes.

4. The integrated design method for hovercraft parameters according to claim 1, characterized in that, The two-stage Bayesian optimization strategy in step S104 specifically includes: a first-stage fine-tuning: based on the basic structural parameters and their corresponding strategy hyperparameters. Construct the initial design vector with the optimal strategy hyperparameters Based on the baseline, a first design space is formed according to the preset search range. Within the first design space, a Bayesian optimization method based on a Gaussian process surrogate model and an expected improvement acquisition function is used to find the preliminary optimal design vector. The second stage is fine-tuning: with the preliminary optimal design vector as the center, the search range of the basic structure parameters and controller hyperparameters is narrowed to form a second design space. Navigation tasks with multiple difficulty levels are constructed for evaluation. Within the second design space, the Bayesian optimization method is used to continue to find the average performance of the navigation tasks with multiple difficulty levels to obtain the global optimal design vector.

5. The integrated design method for hovercraft parameters according to claim 4, characterized in that, The specific steps of the Bayesian optimization method include: Initialization: Within the design space, an initial sample point set is generated using Latin hypercube sampling, and its performance cost is obtained through simulation evaluation, forming the initial observation dataset; Iterative optimization: In each iteration, a Gaussian process model is constructed based on the current observation dataset. By maximizing the expected improvement acquisition function, the next design vector to be evaluated is recommended; Evaluation and update: The design vector to be evaluated is simulated to obtain its performance cost, and the new data points generated by the simulation evaluation are added to the observation dataset to update the Gaussian process model; Convergence judgment: When the maximum number of iterations is reached, or the expected improvement value of the expected improvement acquisition function output is lower than a threshold, or the improvement rate of the performance cost of continuous iterations is lower than a threshold, the iteration terminates, and the design vector with the best performance is selected from the historical data as the output.

6. The integrated design method for hovercraft parameters according to claim 1, characterized in that, The strategy hyperparameter α is a scalar that takes continuous values ​​in the interval (0,1). Through a predefined functional relationship, it continuously adjusts the state error weight matrix and the control input weight matrix of the linear parameter time-varying model predictive controller to achieve a smooth trade-off between preference trajectory tracking performance and preference system robustness.

7. The integrated design method for hovercraft parameters according to claim 5, characterized in that, Also includes: By integrating a data-driven LSTM residual predictor into the LPV-MPC controller, an adaptive controller with online prediction-correction capability is constructed. Based on an adaptive controller, disturbances are introduced into the simulation environment, and the global optimal design vector is corrected during the navigation mission.

8. An integrated parameter design device for hovercraft, characterized in that, include: A random sampling unit is used to fix the basic structural parameters of the hovercraft's propulsion system and, based on these basic structural parameters, randomly sample the hyperparameters of different controller strategies. ; Performance cost calculation unit, used for different strategy hyperparameters The preset navigation task is executed in a simulation environment, and the corresponding performance cost is obtained. ; The adaptive optimization unit is used to analyze policy hyperparameters using an adaptive optimization method based on historical gradient estimation. The trend of changes in system performance is analyzed, and the strategy hyperparameters are updated iteratively accordingly. Until the performance cost After numerical convergence, the optimal policy hyperparameters are obtained. ; The cooperative search unit is used to unfix the infrastructure parameters and combine the infrastructure parameters with the strategy hyperparameters. Together they form a design vector, with the optimal strategy hyperparameters Based on this, a two-stage Bayesian optimization strategy is used to collaboratively search the design vector to obtain the globally optimal design vector. An integrated design unit is used for the integrated structural and control design of the hovercraft based on the optimal design vector.

9. An integrated parameter design device for hovercraft, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be executed by the processor to implement the integrated design method for hovercraft parameters as described in any one of claims 1 to 7.

10. A hovercraft, characterized in that, The design is generated using the integrated parameter design method for hovercraft as described in any one of claims 1 to 7.

Images