Aircraft initial pitch angle acquisition method based on neural network
By combining neural network model with traditional methods, the stability and accuracy of the traditional aircraft's initial pitch angle acquisition method during environmental changes is solved, and efficient aircraft autonomous control and navigation accuracy are achieved.
Patent Information
- Application Number
- CN202510239928.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-03
- Publication Date
- 2025-07-18
AI Technical Summary
Traditional aircraft initial pitch angle acquisition methods are highly dependent on environmental variables, especially when meteorological conditions change rapidly, the stability and accuracy of prediction results are difficult to guarantee, and their ability to express complex nonlinear relationships is limited.
Using the neural network model, the desired range and associated parameters are used as inputs, the pitch angle of the aircraft is predicted through the training data set, and the ballistic model is constructed for iterative optimization. Combined with sensitivity analysis, the sampling interval and step length are adjusted, and compiled into a dynamic link library to integrate iterative prediction in airborne software.
It improves the autonomous control capability and navigation accuracy in the launch stage of the aircraft, can quickly respond to environmental changes, enhances the cross-platform adaptability and portability of the model, and improves prediction accuracy.
Smart Images
Figure CN120337390A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for obtaining the initial pitch angle of an aircraft based on a neural network, belonging to the technical field of aircraft control. Background Art
[0002] As an important parameter in the launch phase, the initial pitch angle directly affects the flight trajectory, hitting accuracy and fuel utilization efficiency of the aircraft.
[0003] The traditional method for obtaining the initial pitch angle of an aircraft is mainly based on classical physical models, and the prediction model is derived by analyzing mechanical equations and aerodynamic characteristics. However, these methods are highly dependent on environmental variables. Especially when meteorological conditions such as temperature and pressure change rapidly, it is difficult to guarantee the stability and accuracy of the prediction results. In addition, the traditional methods have limited ability to express complex non-linear relationships and cannot comprehensively describe the internal laws in the scenario of multi-parameter coupling.
[0004] Therefore, it is necessary to conduct research on the existing problems of the method for obtaining the initial pitch angle of an aircraft to solve the above problems. Summary of the Invention
[0005] In order to overcome the above problems, in-depth research has been carried out, and a method for obtaining the initial pitch angle of an aircraft based on a neural network is proposed, including the following steps:
[0006] S1. Set a neural network model with the expected range and associated parameters as inputs and the launch angle as the output, where the associated parameters refer to external environmental or system variables that can indirectly affect the ballistic performance of the aircraft;
[0007] S2. Collect data to obtain a training set;
[0008] S3. Train the neural network model using the training data set, and use the trained neural network to predict the pitch angle of the aircraft;
[0009] S4. Construct a ballistic model to obtain the calculated range by simulating the aircraft trajectory with the predicted pitch angle;
[0010] S5. Compare the calculated range with the expected range, and determine whether the error is greater than the threshold. If it is greater than the threshold, repeat S2 - S5;
[0011] S6. Use the trained neural network to predict the pitch angle of the aircraft.
[0012] In a preferred embodiment, the associated parameters include wind speed and temperature.
[0013] In a preferred embodiment, the associated parameters further include the target height.
[0014] In a preferred embodiment, the neural network adopts a BP neural network.
[0015] In a preferred embodiment, in S2, the neural network model may be trained multiple times. Before each training, the training set is adjusted.
[0016] In a preferred embodiment, sampling is adjusted according to sensitivity. Through sensitivity analysis, the sampling intervals and / or sampling steps of different variables are adjusted, and data is collected to obtain the training set.
[0017] In a preferred embodiment, in S4, the ballistic model is expressed as:
[0018]
[0019] where m represents the mass of the aircraft, V m represents the velocity of the aircraft, t represents the time constant, P represents the pressure, α represents the angle of attack, β represents the sideslip angle, g represents the acceleration due to gravity, θ represents the ballistic inclination angle, γ v represents the velocity tilt angle, ψ v represents the ballistic deflection angle, X, Y, and Z represent the components of the aerodynamic force in three directions (usually drag, lift, and side force), J x 、J z 、J y represent the moments of inertia of the aircraft about three axes, w x 、w y 、w z represent the angular velocities of the aircraft about three axes, M x 、M y 、M z represent the moment components acting on the aircraft, represents the pitch angle, ψ represents the roll angle, γ represents the roll angle.
[0020] In a preferred embodiment, in S5, the trained neural network is compiled into a dynamic link library and integrated into the aircraft on-board software to predict the pitch angle of the aircraft in real time.
[0021] The beneficial effects of the present invention include:
[0022] (1) Utilize the powerful non-linear modeling ability of the neural network to accurately describe the complex interaction relationships between multiple parameters;
[0023] (2) Optimize the sample design through sensitivity analysis to improve the response ability of the model to changes in input variables;
[0024] (3) Through the design of the dynamic link library, enhance the cross-platform adaptability and portability of the model;
[0025] (4) It can quickly respond to environmental changes, significantly improving the autonomous control ability and navigation accuracy during the aircraft launch phase. Description of the Drawings
[0026] Figure 1 Shows a schematic flow diagram of a method for obtaining the initial pitch angle of an aircraft based on a neural network according to a preferred embodiment of the present invention;
[0027] Figure 2 Shows the simulation trajectory under the initial conditions in the first column of Table 1 in Example 1. Detailed Embodiment
[0028] The present invention will be further described in detail below with reference to the drawings and embodiments. Through these descriptions, the features and advantages of the present invention will become more clear and definite.
[0029] The special term "exemplary" herein means "serving as an example, embodiment, or illustration". Any embodiment described herein as "exemplary" is not necessarily to be construed as superior or better than other embodiments. Although various aspects of the embodiments are shown in the drawings, the drawings do not have to be drawn to scale unless otherwise specified.
[0030] A method for obtaining the initial pitch angle of an aircraft based on a neural network provided by the present invention, as Figure 1 shown, includes the following steps:
[0031] S1. Set a neural network model with the desired range and associated parameters as inputs and the launch angle as the output, where the associated parameters refer to external environments or system variables that can indirectly affect the ballistic performance of the aircraft;
[0032] S2. Collect data to obtain a training set;
[0033] S3. Train the neural network model using the training data set, and use the trained neural network to predict the pitch angle of the aircraft;
[0034] S4. Build a ballistic model, and simulate the aircraft trajectory with the predicted pitch angle to obtain the calculated range;
[0035] S5. Compare the calculated range with the desired range, and determine whether the error is greater than the threshold. If it is greater than the threshold, repeat S2 - S5;
[0036] S6. Use the trained neural network to predict the pitch angle of the aircraft.
[0037] According to the present invention, the use of a neural network makes up for the deficiencies of traditional analytical methods in high - dimensional non - linear problems, making the ballistic design in complex environments more efficient and accurate.
[0038] Furthermore, in the present invention, through steps S1 to S3, the elevation angle is inversely calculated based on the desired range, and based on the result of the inverse calculation, the range is directly calculated through S4 to S5. The directly calculated range result is compared with the inversely calculated desired range, thereby realizing the iterative optimization of the neural network. This method combines the neural network with the theoretical model, which can not only quickly calculate complex ballistic parameters but also improve the prediction accuracy. In addition, during the iterative optimization process, it can comprehensively feedback on many dynamic environmental factors.
[0039] Furthermore, in the present invention, the core of the inverse calculation is the neural network model, and the core of the direct calculation is the theoretical model.
[0040] In S1, the desired range and the elevation angle belong to causal parameters. Causal parameters refer to the core variables that directly affect the flight trajectory and landing point of the aircraft. They act on the ballistic modeling through clear physical mechanisms or dynamic equations.
[0041] The associated parameters refer to external environmental or system variables that can indirectly affect the ballistic performance of the aircraft. They have a certain statistical correlation with indicators such as the range and landing point accuracy of the aircraft, but there is no direct causal relationship. Although the associated parameters do not directly determine the flight trajectory of the aircraft, they indirectly affect the ballistic performance by changing factors such as air density and drag.
[0042] To make full use of the characteristics of the associated parameters and causal parameters, the construction of the data set needs to achieve the organic combination of the two. On the one hand, the change trend of the causal parameters can be predicted using the associated parameters, so as to provide compensation information when the causal parameters are difficult to directly measure or obtain. On the other hand, the statistical correlation and physical interpretability of the associated parameters can be improved through the accurate modeling of the causal parameters.
[0043] Preferably, the associated parameters include wind speed and temperature.
[0044] Preferably, the associated parameters also include the target height.
[0045] Since the air temperature and air pressure are different at different heights and have different effects on the ballistic, accurate analysis of the height can improve the adaptability of the aircraft to different environments.
[0046] In a preferred embodiment, the neural network uses a BP neural network. Preferably, the BP neural network has 5 hidden layers, with 10 neurons in each layer, and the activation function is the ReLU function.
[0047] Preferably, data in the flight environment is collected to construct a training set, and the neural network is trained using the training set.
[0048] In S2, the neural network model may be trained multiple times. Before each training, the training set is adjusted to make the prediction of the neural network more accurate.
[0049] In a preferred embodiment, the data set is obtained through aircraft simulation.
[0050] Preferably, sampling is adjusted according to sensitivity. Through sensitivity analysis, the sampling interval and / or sampling step size are adjusted, and data is collected to obtain the training set.
[0051] The sensitivity analysis is used to analyze the contribution rate of uncertain model input parameters to the uncertainty of the model output. By applying a small perturbation to each input variable one by one and observing the change in the output, the sensitivity of each input variable to the output result is calculated, so as to quantify the strength of the influence of the input variable on the output. For example, when the associated parameters are wind speed and temperature, that is, the inputs of the neural network are expected range, wind speed and temperature, by the method of controlling variables, the other two variables are fixed, and the influence of another variable on the launch angle is analyzed.
[0052] Preferably, the sensitivity is expressed as:
[0053]
[0054] where i represents different input variables, Y p (i) is the launch angle of the neural network output after a small perturbation occurs in the input of the i-th variable, Y t (i) is the launch angle of the neural network output at the reference point input, d is the relative change ratio of the input perturbation, X(i) is the reference value of the input variable i, and S(i) represents the sensitivity of the i-th variable, that is, it represents the change amount of the launch angle when the input variable changes by one unit.
[0055] According to the present invention, the sampling interval and sampling step size are adjusted according to the sensitivities of different variables, so that the data collected each time is different, and the training set samples are more uniform and effective.
[0056] For example, when the obtained temperature sensitivity is 0.1, temperature sampling is performed at a step size of 10°.
[0057] In S4, the ballistic model can describe the motion state of the aircraft under the force environment and its change law, so as to predict the flight trajectory through numerical calculation methods.
[0058] Preferably, the ballistic model is expressed as:
[0059]
[0060] where m represents the mass of the aircraft, V mIndicates the aircraft speed, t represents the time constant, P represents the pressure, α represents the angle of attack, β represents the sideslip angle, g represents the acceleration due to gravity, θ represents the ballistic inclination angle, γ v Indicates the speed tilt angle, ψ v Indicates the ballistic deflection angle, X, Y, and Z represent the components of the aerodynamic force in three directions (usually drag, lift, and side force), J x 、J z 、J y Indicates the moment of inertia of the aircraft about three axes, w x 、w y 、w z Indicates the angular velocity of the aircraft about three axes, M x 、M y 、M z Indicates the moment components acting on the aircraft, Indicates the pitch angle, ψ represents the roll angle, γ represents the roll angle.
[0061] In S5, the specific setting value of the threshold is not limited, and those skilled in the art can freely set it according to actual needs. Preferably, the threshold is set to 0.2%.
[0062] Preferably, in S5, the trained neural network is compiled into a dynamic link library and integrated into the aircraft onboard software to predict the pitch angle of the aircraft in real time.
[0063] The setting of the dynamic link library (DLL) facilitates the calling of it by other software in the onboard computer, is convenient for engineering practice, and enhances the cross-platform adaptability and portability of the model.
[0064] Embodiment
[0065] Embodiment 1
[0066] Conduct an elevation angle prediction experiment for a certain aircraft, including the following steps:
[0067] S1. Set up a neural network model with the desired range and associated parameters as inputs and the elevation angle as the output. The associated parameters refer to external environments or system variables that can indirectly affect the aircraft ballistic performance;
[0068] S2. Collect data to obtain a training set;
[0069] S3. Train the neural network model using the training dataset and predict the pitch angle of the aircraft using the trained neural network;
[0070] S4. Build a ballistic model to simulate the aircraft trajectory with the predicted pitch angle to obtain the calculated range;
[0071] S5. Compare the calculated range with the expected range to determine whether the error is greater than the threshold. If it is greater than the threshold, repeat S2 - S5;
[0072] S6. Use the trained neural network to predict the pitch angle of the aircraft.
[0073] The associated parameters include wind speed, temperature, and target altitude.
[0074] The neural network uses a BP neural network. The BP neural network has 5 hidden layers, with 10 neurons in each layer, and the activation function is the ReLU function.
[0075] In S2, the neural network model may be trained multiple times. Before each training, adjust the training set to make the prediction of the neural network more accurate. The sensitivity is expressed as:
[0076]
[0077] In S4, the trajectory model is expressed as:
[0078]
[0079]
[0080] In S5, compile the trained neural network into a dynamic link library, and set the threshold to 0.2%.
[0081] During the experiment, the expected range, prediction result input into the neural network in S3, and the calculated range in S4 are shown in Table 1.
[0082] Table 1
[0083] Input Initial temperature / ° 300 280 300 Initial pressure / pa 97325.10 97300 90000 Expected range / m 5004.30 4800 4000 Initial launch angle / mil 214.819 209.384 156.457 Verified range / m 5005.87 4796.54 4004.71
[0084] As can be seen from Table 1, under different initial conditions, the error between the expected range input into the neural network and the calculated range in S4 is within 0.2%, indicating that the trained neural network can predict the pitch angle of the aircraft. Under the initial conditions in the first column, the simulation trajectory is as Figure 2 shown.
[0085] The present invention has been described in combination with preferred embodiments above. However, these embodiments are only exemplary and only serve an illustrative purpose. On this basis, various substitutions and improvements can be made to the present invention, and these all fall within the protection scope of the present invention.
Claims
1. A method for obtaining the initial pitch angle of an aircraft based on a neural network, characterized in that, It includes the following steps: S1. Set a neural network model with the expected range and associated parameters as inputs and the launch angle as the output. The associated parameters refer to external environments or system variables that can indirectly affect the ballistic performance of the aircraft. S2. Collect data to obtain a training set. S3. Use the training data set to train the neural network model, and use the trained neural network to predict the pitch angle of the aircraft. S4. Construct a ballistic model, and simulate the aircraft trajectory with the predicted pitch angle to obtain the calculated range. S5. Compare the calculated range with the expected range to determine whether the error is greater than the threshold. If it is greater than the threshold, repeat S2 - S5. S6. Use the trained neural network to predict the pitch angle of the aircraft.
2. The method for obtaining the initial pitch angle of an aircraft based on a neural network according to claim 1, characterized in that the associated parameters include wind speed and temperature.
3. The method for obtaining the initial pitch angle of an aircraft based on a neural network according to claim 2, characterized in that the associated parameters further include the target altitude.
4. The method for obtaining the initial pitch angle of an aircraft based on a neural network according to claim 1, characterized in that the neural network adopts a BP neural network.
5. The method for obtaining the initial pitch angle of an aircraft based on a neural network according to claim 1, characterized in that In S2, the neural network model may be trained multiple times. Before each training, adjust the training set.
6. The method for obtaining the initial pitch angle of an aircraft based on a neural network according to claim 5, characterized in that Adjust the sampling according to the sensitivity. Through sensitivity analysis, adjust the sampling interval and / or sampling step size of different variables, and collect data to obtain the training set.
7. The method for obtaining the initial pitch angle of an aircraft based on a neural network according to claim 6, characterized in that In S4, the ballistic model is expressed as: Among them, m represents the mass of the aircraft, V m represents the velocity of the aircraft, t represents the time constant, P represents the pressure, α represents the angle of attack, β represents the sideslip angle, g represents the acceleration due to gravity, θ represents the ballistic inclination angle, γ v represents the velocity inclination angle, ψ v represents the ballistic deflection angle, X, Y, and Z represent the components of the aerodynamic force in three directions, J x 、J z 、J y represent the moments of inertia of the aircraft about three axes, w x 、w y 、w z represent the angular velocities of the aircraft about three axes, M x 、M y 、M z represent the moment components acting on the aircraft, represents the pitch angle, ψ represents the roll angle, γ represents the roll angle.
8. The method for obtaining the initial pitch angle of an aircraft based on a neural network according to claim 1, characterized in that In S5, compile the trained neural network into a dynamic link library and integrate it into the aircraft on-board software to predict the pitch angle of the aircraft in real time.