Airship flight simulation training method, product, medium and equipment

Abstract

The invention discloses an airship flight simulation training method, product, medium and equipment, and relates to the technical field of airship simulation training, and the method comprises the steps: obtaining flight training data of a stratospheric airship; based on the flight training data, establishing a dynamic mathematical model of the airship by adopting a dynamic sensing method; training a first deep neural network based on the flight training data, and taking the trained first deep neural network as a motion simulation neural network; training the second deep neural network according to the dynamic mathematical model and the motion simulation neural network, and taking the trained second deep neural network as a controller parameter optimization network; performing closed-loop synchronous training on the dynamic mathematical model, the motion simulation neural network and the controller parameter optimization network to obtain a flight simulation model; and carrying out flight simulation training on the stratospheric airship to be trained by adopting the flight simulation model. According to the invention, the comprehensiveness and efficiency of airship simulation training can be effectively improved.

Description

Technical Field

[0001] The present application relates to the technical field of airship simulation training, and in particular to an airship flight simulation training method, product, medium and equipment. Background Art

[0002] As an important aircraft, airships have broad application prospects in military and civilian fields. However, actual flight tests of airships often face problems such as high cost, high risk, and long cycle. Traditional airship flight tests require a lot of manpower, material and financial resources, and are restricted by factors such as weather and airspace, making it difficult to conduct a large number of repeated tests.

[0003] With the development of computer technology and simulation technology, digital simulation has become an important means of aircraft design, testing and training. However, the digital simulation system currently built for airship simulation training still has problems such as single function, low simulation accuracy and poor interactivity, which makes it difficult to meet the needs of comprehensive airship simulation training, resulting in low efficiency of airship simulation training. Therefore, it is urgent to develop a high-precision, multifunctional and highly interactive airship flight simulation training method to improve the efficiency of airship simulation training. Summary of the invention

[0004] The purpose of this application is to provide an airship flight simulation training method, product, medium and equipment, which can effectively improve the comprehensiveness and efficiency of airship simulation training.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides an airship flight simulation training method, comprising:

[0007] Acquiring flight training data of the stratospheric airship; the flight training data includes: position, attitude, speed, acceleration and control input;

[0008] Based on flight training data, a dynamic mathematical model of the airship is established using a dynamic perception method;

[0009] Training a first deep neural network based on the flight training data, and using the trained first deep neural network as a motion imitation neural network;

[0010] The second deep neural network is trained according to the dynamic mathematical model and the motion imitation neural network, and the trained second deep neural network is used as a controller parameter optimization network;

[0011] Conduct closed-loop synchronous training on the dynamic mathematical model, motion imitation neural network and controller parameter optimization network to obtain a flight simulation model;

[0012] A flight simulation model is used to carry out flight simulation training on the stratospheric airship to be trained.

[0013] Optionally, the step of establishing a dynamic mathematical model of the airship based on the flight training data by using a dynamic perception method specifically includes:

[0014] Based on the flight training data, the dynamic mathematical model of the airship established by the dynamic perception method is: Where q is the generalized coordinate vector of the airship; Indicates the first-order derivative operation on q; is the generalized acceleration vector of the airship; F ext is the external force vector of the airship; F control is the control force vector of the airship; M(q) is the generalized mass matrix; is the nonlinear term of the dynamic mathematical model.

[0015] Optionally, the motion imitation neural network takes maximizing a reward function as a training objective; the reward function r t For r t =ω p r p,t +ω attitude r attitude,t +ω traj r traj,t ; Among them, r p,t 、r attitude,t and r traj,t are the position error, attitude error and trajectory tracking error of the airship respectively; ω p ,ω attitude And ω traj are the weights corresponding to position error, attitude error and trajectory tracking error respectively.

[0016] Optionally, the controller parameter optimization network takes minimizing the loss function as the training objective; the loss function loss is Among them, ||·|| represents the two-norm operation; is the second-order derivative of the current state of the airship; M -1 is the inverse matrix of the generalized mass matrix; F env is the correction term for environmental factors; u is the output of the airship controller.

[0017] Optionally, the environmental factor correction term F env F env =f temp (T)×f press (P)×f wind (W); where f temp (T), f press (P) and f wind(W) are the influence functions of temperature T, air pressure P and wind speed W respectively.

[0018] Optionally, the flight simulation model takes maximizing a comprehensive reward function as a training objective; the comprehensive reward function R total For R total =ω1R model +ω2R imitation +ω3R control ; Among them, R model , R imitation and R control are the reward functions of the dynamic mathematical model, the trained motion imitation neural network and the trained controller parameter optimization network; ω1, ω2 and ω3 are the weights corresponding to the reward functions of the dynamic mathematical model, the trained motion imitation neural network and the trained controller parameter optimization network.

[0019] Optionally, after the flight simulation model is used to perform flight simulation training on the stratospheric airship to be trained, the method further includes:

[0020] The controller performance of the stratospheric airship to be trained is evaluated in terms of trajectory tracking error, energy consumption, and stability, and the controller parameters are fine-tuned based on the evaluation results.

[0021] In a second aspect, the present application provides a computer program product, including a computer program, which implements the airship flight simulation training method when executed by a processor.

[0022] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the airship flight simulation training method when executed by a processor.

[0023] In a fourth aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the airship flight simulation training method.

[0024] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0025] The present application provides an airship flight simulation training method, product, medium and equipment, which constructs a high-precision stratospheric digital space through the dynamic mathematical model of the stratospheric airship, the motion simulation neural network and the controller parameter optimization network, and performs closed-loop synchronous training on the dynamic mathematical model, the motion simulation neural network and the controller parameter optimization network, and uses the trained flight simulation model to perform flight simulation training on the stratospheric airship to be trained, thereby realizing all-round simulated flight of the stratospheric airship and effectively improving the comprehensiveness and efficiency of the airship simulation training. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 A schematic diagram of the flow of the airship flight simulation training method provided in this application. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0029] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0030] The process of constructing a flight simulation model in the airship flight simulation training method of the present application mainly includes the modeling process of the dynamic mathematical model of the airship dynamic perception, the data-driven airship motion imitation learning process, the data-driven controller parameter optimization learning process and the closed-loop synchronous training process of the three neural networks, as shown below.

[0031] like Figure 1 As shown, the present application discloses an airship flight simulation training method, comprising:

[0032] Step 1: Obtain flight training data for the stratospheric airship.

[0033] Specifically, through actual flight tests or high-fidelity numerical simulations, flight data of the stratospheric airship is collected as flight training data, including information such as position, attitude, velocity, acceleration, and control input.

[0034] The modeling process of the dynamic mathematical model of the airship dynamic perception is mainly based on the collected flight training data to build the dynamic mathematical model of the airship. The dynamic mathematical model can accurately predict the next state of the airship under a given control input. The specific process is shown in step 2.

[0035] Step 2: Based on the flight training data, a dynamic perception method is used to establish the dynamic mathematical model of the airship.

[0036] Specifically, this application uses a dynamic perception method to perform mechanical modeling on the stratospheric airship and obtains a dynamic mathematical model:

[0037]

[0038] Where q is the generalized coordinate vector of the airship; Indicates the first-order derivative operation on q; is the generalized acceleration vector of the airship; F ext is the external force vector of the airship (including aerodynamic force, gravity, etc.); F control is the control force vector of the airship; M(q) is the generalized mass matrix; It is the nonlinear term of the dynamic mathematical model, specifically represented by Coriolis force and other nonlinear terms.

[0039] In addition, in order to more accurately simulate the impact of the stratospheric environment on the airship, this application introduces the environmental factor correction term F env :

[0040] F env =f temp (T)×f press (P)×f wind (W)(2)

[0041] Among them, f temp (T), f press (P) and f wind (W) are the influence functions of temperature T, air pressure P and wind speed W respectively.

[0042] For the data-driven airship motion imitation learning process, the process uses the collected flight training data to train the first deep neural network, and uses the trained first deep neural network as the motion imitation neural network so that it can accurately reproduce the motion trajectory of the actual airship, as shown in step 3.

[0043] Step 3: Train the first deep neural network based on the flight training data, and use the trained first deep neural network as the motion imitation neural network.

[0044] In order to make the final digital twin model, i.e. the flight simulation model, accurately simulate the movement of the actual airship, this application constructs a motion simulation neural network. The movement mimics a neural network Used to input the status of the airship s m (including position, attitude, speed, etc.), and predict the next state of the airship.

[0045] Motion-mimicking neural networks The training goal is to maximize the reward function r t :

[0046]

[0047] Where t is time, k is the total training time of the motion imitation neural network; J(t) is the total reward function, that is, the maximized reward function; γ t is the weight of the reward function; Express expectations.

[0048] Reward function r t Consists of multiple sub-awards:

[0049] r t =ω p r p,t +ω attitude r attitude,t +ω traj r traj,t (4)

[0050]

[0051] Among them, r p,t 、r attitude,t and r traj,t are the position error, attitude error and trajectory tracking error of the airship respectively; ω p ,ω attitude And ω traj are the weights corresponding to position error, attitude error and trajectory tracking error respectively; exp is the exponential function; P t is the current position of the airship; P desired,t is the desired state of the current position of the airship; σ p is the position state reward parameter; θ t is the current attitude state of the airship; θ desired,t is the expected state of the current attitude of the airship; σ attitude is the posture state reward parameter; traj t The current trajectory tracking status of the airship; traj desired,t Track the desired state for the current trajectory of the airship; σ trajis the trajectory tracking state reward parameter.

[0052] For the data-driven controller parameter optimization learning process, the process optimizes the parameters of the airship controller by constructing a neural network structure, namely, a controller parameter optimization network. The controller parameter optimization network receives the current airship state and the desired state as input and outputs the adjustment value of the controller parameter, as shown in step 4.

[0053] Step 4: Train the second deep neural network according to the dynamic mathematical model and the motion imitation neural network, and use the trained second deep neural network as the controller parameter optimization network.

[0054] Based on the dynamic mathematical model of the airship and the motion simulation neural network, this application constructs a controller parameter optimization network π θ (a), the controller parameter optimization network π θ (a) Receives the desired airship state as input and outputs the controller parameter adjustment value u:

[0055]

[0056] Among them, K p , K d , and K i are the proportional, differential and integral gain matrices respectively; x d and x are the desired state and the actual state respectively; and x are the desired states x d and the first-order derivative of the actual state x.

[0057] Controller parameter optimization network π θ The training goal of (a) is to minimize the following loss function loss:

[0058]

[0059] Among them, ||·|| represents the two-norm operation; is the second-order derivative of the current state of the airship; M -1 is the inverse matrix of the generalized mass matrix; F env is the environmental factor correction term; u is the output of the airship controller. The loss function loss includes the state prediction error (the first two norms) and the energy consumption of the control input (the second two norms).

[0060] For the closed-loop synchronous training process of three neural networks, in order to achieve efficient controller parameter adjustment, this application proposes a closed-loop synchronous training method of three neural networks. The three neural networks are a dynamic mathematical model, a motion imitation neural network, and a controller parameter optimization network. The dynamic mathematical model of this application is also composed of a deep neural network. This process performs closed-loop synchronous training on the three neural networks by using a defined comprehensive reward function. During the training process, the parameters of each network are continuously adjusted to maximize the comprehensive reward, as shown in step 5.

[0061] Step 5: Perform closed-loop synchronous training on the dynamic mathematical model, motion simulation neural network and controller parameter optimization network to obtain a flight simulation model.

[0062] The process of implementing closed-loop synchronous training adopts an alternating optimization strategy, and the training of each network takes into account the outputs of the other two networks to ensure the consistency and stability of the entire system consisting of the dynamic mathematical model, the motion imitation neural network, and the controller parameter optimization network. Specifically, the application first establishes a dynamic mathematical model, then performs an outer loop simulation through the motion imitation neural network, then adjusts the parameters of the airship controller through the controller parameter optimization network, and then enters the dynamic mathematical model, and repeats the cycle.

[0063] Specifically, the comprehensive reward function R defined in this application is total :

[0064] R total =ω1R model +ω2R imitation +ω3R control (10)

[0065] Among them, R model , R imitation and R control are the reward functions of the dynamic mathematical model, the trained motion imitation neural network, and the trained controller parameter optimization network; ω1, ω2, and ω3 are the weights corresponding to the reward functions of the dynamic mathematical model, the trained motion imitation neural network, and the trained controller parameter optimization network. imitation That is, J(t), R control That is loss, R model That is, the error between the flight data output by the dynamic mathematical model and the actual flight data.

[0066] The closed-loop synchronous training of the present application ensures the coordinated optimization among the three neural networks, thereby obtaining a high-performance and stable airship control system, namely, a flight simulation model.

[0067] Step 6: Use the flight simulation model to carry out flight simulation training on the stratospheric airship to be trained.

[0068] After obtaining the flight simulation model through the above steps, the flight simulation model is used to perform flight simulation training on the stratospheric airship to be trained. During the flight simulation training, the flight simulation model is used to evaluate the performance of the controller in the digital twin environment. The evaluation indicators include trajectory tracking error, energy consumption, stability, etc. When the tracking error is small, the energy consumption is low, and the stability is strong, the controller parameters are well selected. If the evaluation result is poor, the controller parameters are fine-tuned according to the evaluation result, and the training, evaluation, and fine-tuning process are repeated until a stable state is reached. The optimized controller parameters can also be applied to the actual stratospheric airship for actual flight testing to verify its performance and reliability.

[0069] Compared with the prior art, the airship flight simulation training method of the present application has the following advantages:

[0070] 1. The airship flight simulation training method of the present application greatly reduces the demand for actual flight testing and reduces development costs and risks.

[0071] 2. The airship flight simulation training method of the present application improves the efficiency and accuracy of controller parameter optimization.

[0072] 3. The airship flight simulation training method of the present application realizes accurate simulation of the stratospheric environment through digital twin technology, thereby improving the robustness of the flight simulation model.

[0073] 4. The closed-loop synchronous training in the airship flight simulation training method of the present application ensures the coordinated optimization between various networks and improves the overall performance.

[0074] 5. The airship flight simulation training method of the present application has good scalability, can more easily adapt to different types of stratospheric airships and mission requirements, and can significantly reduce the cost of airship testing: through digital space simulation, a large number of flight tests can be carried out without actual flight, thereby significantly reducing the cost of testing.

[0075] In summary, this application uses digital twin technology to accurately simulate the stratospheric environment and establish a highly simulated digital twin airship model (flight simulation model), which reduces the actual flight test cost, thereby reducing the cost of airship research and development and training, effectively improving the efficiency of airship parameter adjustment training, and accelerating the iteration speed of airship design optimization. This application can achieve high-precision simulation of airship flight, and achieves efficient and accurate controller parameter optimization by combining the modeling process of the dynamic mathematical model of the airship dynamic perception, the data-driven airship motion imitation learning process, the data-driven controller parameter optimization learning process, and the closed-loop synchronous training process of the three neural networks. This method not only greatly reduces the demand for actual flight tests, but also improves the performance and reliability of the control system, and provides an innovative solution for the development and optimization of stratospheric airship controllers, which is of great significance to promoting the development of stratospheric airship technology.

[0076] In some embodiments, the present application also provides a computer program product, including a computer program, which implements the airship flight simulation training method when executed by a processor.

[0077] In some embodiments, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the airship flight simulation training method is implemented.

[0078] In some embodiments, the present application also provides a computer device, including a processor, a memory, an input / output interface (Input / Output, referred to as I / O), a communication interface, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the airship flight simulation training method.

[0079] The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the airship flight simulation training method is implemented.

[0080] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0081] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ReadOnlyMemory, ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0082] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0083] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. An airship flight simulation training method, characterized in that: include: Obtain flight training data for stratospheric airships; The flight training data includes: position, attitude, speed, acceleration and control input; Based on flight training data, a dynamic mathematical model of the airship is established using a dynamic perception method; Training a first deep neural network based on the flight training data, and using the trained first deep neural network as a motion imitation neural network; The second deep neural network is trained according to the dynamic mathematical model and the motion imitation neural network, and the trained second deep neural network is used as a controller parameter optimization network; Conduct closed-loop synchronous training on the dynamic mathematical model, motion imitation neural network and controller parameter optimization network to obtain a flight simulation model; A flight simulation model is used to carry out flight simulation training on the stratospheric airship to be trained.

2. The airship flight simulation training method according to claim 1, characterized in that: The dynamic mathematical model of the airship is established based on the flight training data by using a dynamic perception method, which specifically includes: Based on the flight training data, the dynamic mathematical model of the airship established by the dynamic perception method is: Where q is the generalized coordinate vector of the airship; Indicates the first-order derivative operation on q; is the generalized acceleration vector of the airship; F ext is the external force vector of the airship; F control is the control force vector of the airship; M(q) is the generalized mass matrix; is the nonlinear term of the dynamic mathematical model.

3. The airship flight simulation training method according to claim 1, characterized in that: The motion imitation neural network takes maximizing the reward function as the training goal; the reward function r t For r t =ω p r p,t +ω attitude r attitude,t +ω traj r traj,t ; Among them, r p,t 、r attitude,t and r traj,t are the position error, attitude error and trajectory tracking error of the airship respectively; ω p ,ω attitude And ω traj are the weights corresponding to position error, attitude error and trajectory tracking error respectively.

4. The airship flight simulation training method according to claim 2, characterized in that: The controller parameter optimization network takes minimizing the loss function as the training objective; the loss function loss is Among them, ||·|| represents the two-norm operation; is the second-order derivative of the current state of the airship; M -1 is the inverse matrix of the generalized mass matrix; F env is the correction term for environmental factors; u is the output of the airship controller.

5. The airship flight simulation training method according to claim 2, characterized in that: The environmental factor correction term F env F env =f temp (T)×f press (P)×f wind (W); where f temp (T), f press (P) and f wind (W) are the influence functions of temperature T, air pressure P and wind speed W respectively.

6. The airship flight simulation training method according to claim 1, characterized in that: The flight simulation model takes maximizing the comprehensive reward function as the training goal; the comprehensive reward function R total For R total =ω1R model +ω2R imitation +ω3R control ; Among them, R model , R imitation and R control are the reward functions of the dynamic mathematical model, the trained motion imitation neural network and the trained controller parameter optimization network; ω1, ω2 and ω3 are the weights corresponding to the reward functions of the dynamic mathematical model, the trained motion imitation neural network and the trained controller parameter optimization network.

7. The airship flight simulation training method according to claim 1, characterized in that: After the flight simulation model is used to perform flight simulation training on the stratospheric airship to be trained, the method further includes: The controller performance of the stratospheric airship to be trained is evaluated in terms of trajectory tracking error, energy consumption, and stability, and the controller parameters are fine-tuned based on the evaluation results.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the airship flight simulation training method according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the airship flight simulation training method according to any one of claims 1 to 7 is implemented.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the airship flight simulation training method according to any one of claims 1 to 7.

Images