Aircraft trajectory rapid generation method based on artificial neural network
Through the rapid generation method of aircraft trajectory based on artificial neural network, the problems of large calculations, weak timeliness and insufficient scalability caused by traditional numerical model driving methods in massive simulations are solved, and efficient calculation and scalability of aircraft trajectory generation are achieved.
Patent Information
- Application Number
- CN202510226326.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-27
- Publication Date
- 2025-06-24
AI Technical Summary
In the current technology, traditional numerical model driving methods in massive simulations have led to the problems of large calculations of aircraft trajectory generation, weak timeliness and insufficient scalability.
A rapid vehicle trajectory generation method based on artificial neural network is adopted to determine the input parameters and output parameters of the mathematical model of the aircraft trajectory feature, and the predicted vehicle trajectory data are generated through training and verification.
It significantly reduces the timeliness of the aircraft trajectory simulation calculation process, improves the calculation efficiency, and has good scalability and universal applicability.
Smart Images
Figure CN120197472A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for quickly generating an aircraft trajectory based on an artificial neural network, belonging to the field of artificial intelligence modeling. Background Art
[0002] The simulation of aircraft trajectory generation is a complex computational process, involving knowledge and technologies of multiple disciplines. The current traditional method for generating an aircraft trajectory is to solve it using the dynamic model and kinematic model of the aircraft. The simulation process involves complex calculations in aspects such as the aerodynamic model and the navigation and guidance model. A small step size during numerical integration and a complex model will lead to a significant increase in the computational amount, while conversely, it is likely to cause problems with poor calculation accuracy of the aircraft trajectory. Especially for a large number of simulation working conditions, the computational timeliness of the numerical model-driven aircraft trajectory generation and solution is particularly crucial. Currently, in a large number of simulations, the traditional numerical model-driven method has problems such as a relatively large simulation computational amount caused by the solution, relatively weak timeliness, and insufficient scalability. Summary of the Invention
[0003] The technical problem to be solved by the present invention is: overcoming the deficiencies of the prior art, providing a method for quickly generating an aircraft trajectory based on an artificial neural network, and solving the problems of relatively large simulation computational amount, relatively weak timeliness, and insufficient scalability caused by the solution of the traditional numerical model-driven method in a large number of simulations.
[0004] The technical solution of the present invention is: In the first aspect, a method for quickly generating an aircraft trajectory based on an artificial neural network is provided, including:
[0005] S1. Determine the input parameters and model output parameters of the mathematical model of the aircraft trajectory characteristics, and determine the range of the input parameters; the model includes the kinematic equation and kinematic constraints established for the aircraft trajectory.
[0006] S2. Based on the numerical calculation results of the mathematical model of the aircraft trajectory characteristics and the range of the input parameters, obtain a simulation sample data set; divide the simulation sample data set into a training sample data set and a test sample data set.
[0007] S3. Construct a neural network model suitable for aircraft trajectory generation; train the neural network model based on the training sample data set to obtain a trained neural network model.
[0008] S4. Verify the trained neural network model based on the test sample data set. If the verification fails, return to S3 to retrain the neural network model; after passing the verification, obtain an accurate neural network model that has been verified.
[0009] S5. Input data within the range of the input parameters into the accurate neural network model that has been verified, and the neural network model can generate and output predicted aircraft trajectory data.
[0010] When the predicted aircraft trajectory data output does not meet the requirements, expand the training sample data set to generate an expanded training sample data set, and return to S2 to iterate again based on the expanded training sample set to obtain a neural network model with the performance of new samples. When inputting data within its parameter range, it can generate and output the predicted aircraft trajectory data.
[0011] Preferably, when obtaining the simulation sample data set, the sample data screening principle is:
[0012] The sample data can cover the entire flight process of the aircraft;
[0013] The sample data can cover the regular characteristics of various working conditions of the aircraft trajectory;
[0014] Under the condition of ensuring the sample quality and uniformity, determine the sample data scale according to the actual operation time requirements.
[0015] Preferably, the neural network model can select any one of the following:
[0016] Radial basis neural network, BP neural network, convolutional neural network.
[0017] Preferably, when training the neural network model based on the training sample data set:
[0018] S2-1. Initialize the parameters in the neural network model, including: neural network weights and activation function thresholds;
[0019] S2-2. Input the training sample data set into the neural network model, and the neural network model outputs the fitted aircraft trajectory;
[0020] S2-3. Compare the fitted aircraft trajectory output by the neural network model with the simulation calculation value of the aircraft trajectory mathematical model to obtain the error between the two;
[0021] S2-4. When the error does not meet the requirements, optimize and adjust the parameters in the neural network model, and return to S2-2 to iterate again; until the error meets the requirements, output the completed neural network model.
[0022] Preferably, the error threshold during training is set to 10%, and it can also be adjusted according to actual needs.
[0023] Preferably, when verifying the trained neural network model based on the test sample data set, the verification passing standard is:
[0024] The fitted aircraft trajectory output by the neural network model is compared with the simulation calculation value of the aircraft trajectory mathematical model. If the error between the two is less than or equal to the error threshold, it is considered verified; the error threshold can be adjusted according to actual requirements.
[0025] Preferably, there are two ways to generate an expanded training sample data set:
[0026] Based on the established kinematic equations and kinematic constraints, the input parameters and model output parameters of the aircraft trajectory feature mathematical model are determined through calculation and merged into the training sample data set for expansion;
[0027] Based on the actual test conditions and parameters, a test sample data set is generated and merged into the training sample data set for expansion.
[0028] Preferably, the method of generating a test sample data set based on the actual test conditions and parameters and merging it into the training sample data set for expansion is as follows:
[0029] The actual test parameters are input into the verified accurate neural network model to generate a predicted aircraft trajectory, which is compared with the aircraft trajectory in the actual test. If the error obtained from the comparison exceeds the error threshold, multiple flight tests will be carried out, and the test sample data set will be generated from the actual test parameters and the actual aircraft trajectory and merged into the training sample data set.
[0030] In a second aspect, a terminal device is provided, including:
[0031] A memory for storing instructions executed by at least one processor;
[0032] A processor for executing the instructions stored in the memory to implement the method for quickly generating an aircraft trajectory based on an artificial neural network as described above.
[0033] In a third aspect, a computer-readable storage medium is provided. The computer-readable storage medium stores computer instructions, and when the computer instructions are run on a computer, the computer is caused to execute the method for quickly generating an aircraft trajectory based on an artificial neural network as described above.
[0034] The present invention has the following advantages compared with the prior art:
[0035] (1) Based on the data-driven neural network modeling method, the performance of the aircraft data model can be expanded by introducing new sample sets as needed at any time.
[0036] (2) The data-driven neural network model adopts a data representation method, which can significantly reduce the timeliness problem in the aircraft trajectory simulation calculation process.
[0037] (3) The neural network model based on data-driven can fit and model the trajectories of the same type of aircraft, enabling a single model to predict the trajectories of multiple aircraft of the same type, with certain general applicability. Description of the Drawings
[0038] Figure 1 Schematic diagram of the multi-input multi-output neural network model structure for the aircraft trajectory of the present invention;
[0039] Figure 2 Schematic diagram of the fitting training process of the neural network model for the aircraft trajectory of the present invention;
[0040] Figure 3 Schematic diagram of the training and construction of the neural network model for the aircraft trajectory of the present invention. Detailed Description of the Invention
[0041] With the breakthrough development of the basic theory of artificial intelligence, artificial intelligence has shown significant advantages and potential application prospects in fields such as aerospace and automatic control with its powerful learning and adaptability. In the face of the rapid generation requirements of aircraft trajectories such as online planning of massive working condition trajectories and exploration of maneuverability boundary, the present invention proposes a data-driven neural network aircraft trajectory generation and modeling method to quickly generate aircraft trajectories and solve the timeliness problem caused by the large computational amount of the numerical model-driven aircraft trajectory under continuous massive simulation conditions.
[0042] To make the objectives, technical solutions and advantages of the present invention clearer, the following further elaborates in detail a method for quickly generating aircraft trajectories based on an artificial neural network proposed by the present invention in conjunction with the drawings and specific embodiments.
[0043] The main workflow of the aircraft trajectory generation method based on neural network model prediction includes: digital characterization of the real physical characteristics of the aircraft trajectory, construction of the aircraft trajectory data sample set, fitting of the aircraft trajectory based on the neural network, and online generation and prediction of the aircraft trajectory model.
[0044] (1) The digital characterization of the real physical characteristics of the aircraft trajectory refers to the parameterization of the aircraft physical characteristics and the trajectory, establishing a numerical model representation between the input quantity and the output quantity, and realizing the parametric digitization of the aircraft physical characteristics.
[0045] (2) The construction of the aircraft trajectory data sample set refers to the set of input data samples and output data samples after the parameterization of the aircraft physical characteristics and the trajectory. According to the functional requirements, the data samples are mainly classified into training sample data sets, test sample data sets, experimental sample data sets, etc., and the data samples are sourced from simulation calculations and experiments.
[0046] Among them, the training sample data set refers to the sample data used for the fitting training of the neural network model in the input data sample and output data sample sets;
[0047] The test sample data set refers to the sample data used for the validity verification of the neural network model in the input data sample and output data sample sets;
[0048] The experimental sample data set refers to the sample data obtained through experiments. If the experimental sample data set is obtained subsequently, it can be used to compare the data generated by the neural network model with the experimental data to test the authenticity of the neural network model. At the same time, according to the quality of the test results, the experimental sample data set can be used as part of the training sample data set to retrain the aircraft trajectory neural network model until an acceptable aircraft trajectory neural network model is obtained.
[0049] (3) The aircraft trajectory fitting based on the neural network refers to using the neural network algorithm to fit the data sample set according to the data sample set of the real physical characteristic parameters, and establishing a mapping relationship neural network model between the aircraft characteristic parameters and the input-output quantities of the trajectory.
[0050] Among them, the neural network algorithm refers to common neural networks such as RNN (Recurrent Neural Networks), CNN (Convolutional Neural Networks), and BP (Back Propagation) neural networks.
[0051] (4) The online generation and prediction of the aircraft trajectory model means that the aircraft trajectory neural network model obtained after neural network fitting can output good prediction results after the data outside the given sample set.
[0052] Among them, the online generation means that the aircraft trajectory neural network model after fitting can quickly output the prediction results outside the sample points.
[0053] In view of the extremely urgent need for real-time performance in the massive simulation calculation of aircraft trajectories or online trajectory calculation. Different from the function model fitting method, the artificial neural network can adaptively update its internal structure based on external data, without first clarifying the complex mapping relationship between the input and output. By training and learning certain rules, the expected output quantity can be obtained. The present invention proposes a method for quickly generating an aircraft trajectory based on artificial neural network fitting and prediction. By using a method combining a large amount of sample data driving and neural network fitting, the problem of long model calculation time can be avoided. It mainly includes the following steps:
[0054] Step 1: Characterization of the input-output parameters of the aircraft trajectory model
[0055] Parametric characterization is a method of describing the characteristics of a physical system using model parameters, as shown in Figure 3 Figure ①. First, the flight trajectory characteristics of the aircraft are described in the form of a mathematical model; secondly, the input parameters and model output parameters of the mathematical model of the aircraft flight trajectory characteristics are determined, and the range of the input parameters is determined.
[0056] Step 2: Screening and setting of the aircraft flight trajectory sample data set
[0057] Based on the numerical calculation results of the mathematical model of the aircraft flight trajectory characteristics and the range of input parameters in Step 1, an input-output simulation sample data set is obtained, as shown in Figure 3 Figure ②.
[0058] First, the sample data screening ensures correctness and accuracy, and can reflect the internal laws of the entire process of the aircraft flight trajectory; secondly, the sample data uniformity can cover the regular characteristics of various working conditions of the aircraft flight trajectory; finally, the sample quantity rationality determines the scale of the sample data according to the actual operation time requirements under the condition of ensuring the sample quality and uniformity.
[0059] 90% of the obtained aircraft flight trajectory simulation sample data set (the percentage can be set according to actual needs) is used as the training sample data set of the neural network model, and 10% of the simulation sample data set (the percentage can be set according to actual needs) is used as the test sample data set of the neural network model.
[0060] Step 3: Construction and training of the aircraft flight trajectory neural network model
[0061] As an operation model, a neural network is composed of a large number of nodes and connections between them. Each node represents a specific output function (activation function), and the connection between each two nodes represents a weighted value (weight) for the signal passing through this connection, as shown in Figure 1 Figure ③.
[0062] The neural network model is trained using the aircraft flight trajectory training sample set obtained in Step 2, as shown in Figure 3 Figure ③. First, a neural network model is selected, such as a radial basis function (RBF) neural network, a BP (BackPropagation) neural network, a convolutional neural network (CNN: Convolutional Neural Networks), etc.; secondly, in combination with the aircraft flight trajectory training sample set data, weight values, activation functions, etc., the fitting training construction of the neural network model is realized. Finally, the successfully fitted aircraft flight trajectory neural network model can better approximate the simulation calculation value of the aircraft flight trajectory mathematical model.
[0063] Step 4: Verification of the aircraft flight trajectory neural network model
[0064] Use the aircraft trajectory test sample data set obtained in Step 2 to verify the prediction performance of the aircraft trajectory neural network model obtained in Step 3, as Figure 3 shown in ④ below.
[0065] First, the input parameters of the test sample data set are given to the aircraft trajectory neural network model, and the output data set of the neural network model is quickly calculated; secondly, the output parameters of the test sample data set are compared with the output data set of the neural network model to verify the accuracy of the quick calculation of the neural network model.
[0066] Generally, if the relative error of the comparison is within 10% (the minimum allowable error can be set according to actual needs), it can be considered that the neural network model can better represent the characteristics of the aircraft trajectory. If the relative error is greater than 10% (the minimum allowable error can be set according to actual needs), otherwise repeat Step 3 and Step 4 until the relative error meets the set requirements, as Figure 2 shown below.
[0067] Step 5: Application of the aircraft trajectory neural network model
[0068] Based on the aircraft trajectory neural network model that has been verified to be relatively accurate obtained in Step 4, if the input parameter data of the given aircraft trajectory (within the parameter range of the sample data set) is provided, the aircraft trajectory neural network model can quickly generate and output relatively accurate aircraft trajectory data. At the same time, it can also predict the output outside the parameter range of the aircraft trajectory sample data set.
[0069] Step 6: Expansion of the sample set of the aircraft neural network model
[0070] If the training sample data set needs to be expanded to generate an expanded training sample data set and then reconstruct the neural network model, Steps 2, 3, 4, and 5 need to be repeated, and a neural network model with the performance of the new samples can be obtained. The retrained neural network model has a wider applicability. For example, fitting models for multiple similar aircraft trajectory sample sets to achieve the prediction of multiple similar aircraft trajectories by a single neural network model.
[0071] There are two ways to generate an expanded training sample data set:
[0072] Based on the established kinematic equations and kinematic constraints, determine the input parameters and model output parameters of the aircraft trajectory characteristic mathematical model through calculation, and merge them into the training sample data set for expansion;
[0073] Based on the actual test conditions and parameters, generate a test sample data set and merge it into the training sample data set for expansion.
[0074] Based on the actual test conditions and parameters, the method for generating a test sample data set and merging it into the training sample data set for expansion is as follows:
[0075] Input the actual test parameters into the verified accurate neural network model to generate a predicted flight trajectory of the aircraft. Compare it with the flight trajectory of the aircraft in the actual test. If the error obtained from the comparison exceeds the error threshold, multiple flight tests will be carried out, and the actual test parameters and the actual flight trajectory of the aircraft will be used to generate a test sample data set, which will be merged into the training sample data set.
[0076] In summary, the present invention discloses a method for quickly generating an aircraft trajectory based on an artificial neural network. Through the data-driven neural network model fitting method, the aircraft trajectory sample data is clustered, and neural network fitting is performed based on different sample data to obtain a data-driven aircraft trajectory neural network model.
[0077] In a second aspect, the present invention provides a terminal device, including:
[0078] A memory for storing instructions executed by at least one processor;
[0079] A processor for executing the instructions stored in the memory to implement the method for quickly generating an aircraft trajectory based on an artificial neural network as described above.
[0080] In a third aspect, the present invention provides a computer-readable storage medium storing computer instructions, which when run on a computer, cause the computer to execute the method for quickly generating an aircraft trajectory based on an artificial neural network as described above.
[0081] Although the present invention has been disclosed above with preferred embodiments, it is not used to limit the present invention. Any person skilled in the art can make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.
[0082] The content not detailedly described in the specification of the present invention belongs to the prior art well-known to those skilled in the art.
Claims
1. A method for rapidly generating aircraft trajectories based on artificial neural networks, characterized in that include: S1. Determine input parameters and model output parameters of a mathematical model of aircraft trajectory characteristics, and determine the range of the input parameters; the model includes kinematic equations and kinematic constraints established for the aircraft trajectory; S2. Based on the numerical calculation results of the mathematical model of the aircraft trajectory characteristics and the input parameter range, a simulation sample data set is obtained; and the simulation sample data set is divided into a training sample data set and a test sample data set; S3, constructing a neural network model suitable for aircraft trajectory generation; Training a neural network model based on a training sample data set to obtain a trained neural network model; S4, verifying the trained neural network model based on the test sample data set. If the verification fails, return to S3 to retrain the neural network model. After the verification passes, an accurate neural network model is obtained. S5. Inputting data within a parameter range into the verified accurate neural network model, the neural network model can generate and output predicted aircraft trajectory data; When the output predicted aircraft trajectory data does not meet the requirements, the training sample data set is expanded to generate an expanded training sample data set, and S2 is returned to iterate again based on the expanded training sample set to obtain a neural network model with the performance of the newly added samples. When data within the parameter range is input into it, the predicted aircraft trajectory data can be generated and output.
2. The method for rapidly generating aircraft trajectories based on artificial neural networks according to claim 1, characterized in that: When obtaining a simulation sample data set, the sample data screening principle is: The sample data can cover the entire flight trajectory of the aircraft; The sample data can cover the regular characteristics of various working conditions of the aircraft trajectory; Under the condition of ensuring sample quality and uniformity, the sample data size is determined according to the actual computing time requirements.
3. The method for rapidly generating aircraft trajectories based on artificial neural networks according to claim 1, characterized in that: The neural network model can choose any of the following: Radial basis neural network, BP neural network, convolutional neural network.
4. The method for rapidly generating aircraft trajectories based on artificial neural networks according to claim 1, characterized in that: When training a neural network model based on a training sample dataset: S2-1, initializing the parameters in the neural network model, including: neural network weights and activation function thresholds; S2-2, inputting a training sample data set into the neural network model, and the neural network model outputs a fitted aircraft trajectory; S2-3, comparing the fitted aircraft trajectory output by the neural network model with the aircraft trajectory mathematical model simulation calculation value to obtain the error between the two; S2-4. When the error does not meet the requirements, optimize and adjust the parameters in the neural network model, return to S2-2 and iterate again; until the error meets the requirements, output the completed neural network model.
5. The method for rapidly generating aircraft trajectories based on artificial neural networks according to claim 4, characterized in that: The error threshold during training is set to 10%, which can also be adjusted according to actual needs.
6. The method for rapidly generating aircraft trajectories based on artificial neural networks according to claim 5, characterized in that: When verifying the trained neural network model based on the test sample data set, the verification pass criteria are: The fitted aircraft trajectory output by the neural network model is compared with the simulated calculation value of the aircraft trajectory mathematical model. If the error between the two is less than or equal to the error threshold, the verification is passed; the error threshold can be adjusted according to actual needs.
7. The method for rapidly generating aircraft trajectories based on artificial neural networks according to claim 1, characterized in that: There are two ways to generate an expanded training sample dataset: Based on the established kinematic equations and kinematic constraints, the input parameters and model output parameters of the mathematical model of the aircraft trajectory characteristics are determined by calculation and merged into the training sample data set for expansion; Based on the actual test conditions and parameters, a test sample data set is generated and merged into the training sample data set for expansion.
8. The method for rapidly generating aircraft trajectories based on artificial neural networks according to claim 7, characterized in that: Based on the actual test conditions and parameters, the test sample data set is generated and merged into the training sample data set for expansion: The actual test parameters are input into the verified accurate neural network model to generate the predicted aircraft trajectory, which is then compared with the aircraft trajectory in the actual test. If the error obtained from the comparison exceeds the error threshold, multiple flight tests will be conducted, and the test sample data set generated by the actual test parameters and the actual aircraft trajectory will be merged into the training sample data set.
9. A terminal device, characterized in that: include: a memory for storing instructions executed by at least one processor; A processor, configured to execute instructions stored in a memory to implement a method according to any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed on a computer, the computer is enabled to execute the method according to any one of claims 1 to 8.