Online intelligent self-adaptive parameter adjusting method for missile stability controller
Through the online intelligent adaptive parameter adjustment method, neural networks are used to perform missile stability perception and parameter adaptive adjustment, which solves the problem of insufficient stability margin in complex flight missions, and achieves the stability control and robustness of missiles in complex environments.
Patent Information
- Application Number
- CN202510833215.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-20
- Publication Date
- 2025-08-26
AI Technical Summary
Traditional missile controllers are unable to effectively deal with model misalignment and uncertainty in complex flight missions, resulting in insufficient stability margins, resulting in a decrease in control divergence and tracking accuracy.
The online intelligent adaptive parameter adjustment method is adopted to construct a missile nonlinear dynamic model, use neural networks to perform stability perception and parameter adaptive adjustment, and design controller parameter adaptive adjustment strategies to achieve stable control of missiles in complex environments.
It realizes stable control of missiles in complex flight environments, improves the robustness and adaptive adjustment capabilities of the controller, and ensures that the system can remain stable when the preset stability margin is insufficient.
Smart Images
Figure CN120540100A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of missile stabilization and control, and relates to a missile stabilization and control method, in particular to an online intelligent adaptive control method. Background Art
[0002] The increasing demands placed on new-generation missiles for complex missions such as cross-domain flight, multi-missile coordination, and intelligent variable-shape flight have led to increasingly severe control challenges in actual flight, including strong model nonlinearity, fast-changing parameters, and strong channel coupling. This has dramatically increased the control difficulty. Currently, traditional controller design relies on ground-based wind tunnel or computational fluid dynamics (CFD) modeling data. However, due to limitations in technical data and testing methods, a large number of unmodeled conditions exist in actual flight. For example, conditions such as missile boundary vortex shedding exciting missile body vibrations cannot be identified on the ground. Flow characteristics such as aerodynamic heating, thin shock layers, viscous interference, and real gas effects during flight are difficult to accurately model. Consequently, deviations and uncertainties are inevitable between the accuracy of models derived from ground testing and those used in actual flight. In actual flight, when model deviations and unforeseen uncertainties caused by complex missions are encountered, inadequate stability margins designed offline under certain operating conditions can easily lead to poor vehicle response, reduced tracking accuracy, and control divergence.
[0003] Traditional full-airspace gain scheduling methods based on linear fitting are no longer able to meet the control performance requirements under conditions of model inaccuracy and uncertainty. The next generation of missiles must be able to intelligently and adaptively adjust to changes in their structural characteristics, handle unpredictable nonlinear changes in aerodynamic parameters and mass characteristics, and maintain stable control despite large, unknown dynamic changes. These are pressing challenges in missile stability control. Summary of the Invention
[0004] In order to solve the above problems and overcome the shortcomings of the existing technology, the present invention provides an online intelligent adaptive parameter adjustment method for a missile stabilization controller, which improves the intelligent adaptive adjustment capability of the missile in actual flight and achieves stable control even when the missile model parameters undergo large changes.
[0005] The specific technical solutions of the present invention are as follows:
[0006] A method for online intelligent adaptive parameter adjustment of a missile stabilization controller comprises the following steps:
[0007] S1. Construct a numerical simulation model of the missile's nonlinear dynamics model; deflect important parameters that affect system stability in different combinations to obtain the state output of the missile's dynamics system under different deflection combinations;
[0008] S2. Construct training data labels by quantifying the stable state of the missile stabilization and control system;
[0009] S3, extracting frequency domain features from the training input data;
[0010] S4. Map the missile system's stability to the rudder command timing signal and construct a neural network for steady-state perception.
[0011] S5. Design a controller parameter adaptive adjustment strategy.
[0012] In the step S1:
[0013] Important parameters that affect system stability include rudder efficiency, center of mass, dynamic coefficient, and dynamic pressure;
[0014] In different pull-bias combinations, the system output includes stable output, unstable output, and output divergence;
[0015] By expanding the bias combination, the sample diversity of the training data can be enriched.
[0016] The step S2 comprises the following steps:
[0017] S21. At each observation data point, the amplitude stability margin of the system at that moment is calculated according to the linear system stability method. The comprehensive stability state of the entire observation data is obtained by the time weighting method. This is used as the system stability state metric at the current moment. The calculation is expressed as follows:
[0018]
[0019] Where △ represents the system stability at the current moment, N represents the length of the observation segment data, k represents the number of simulation steps from the start of simulation to the current moment, and k ≥ N, γ represents the attenuation coefficient, γ = 0.995, PGM represents the positive amplitude stability margin, and NGM represents the negative amplitude stability margin.
[0020] S22, the system stable state is represented by a value between [-1, 1];
[0021] S23, classifying the missile stability state into different stability levels according to different instability levels, and obtaining time domain state characteristic data under different instability conditions;
[0022] The stability level classification includes 5 levels, specifically divided into I negative instability, II insufficient negative stability margin, III stable, IV insufficient positive stability margin, and V positive instability. The specific classification standard is shown in formula (2):
[0023]
[0024] Among them: σ1, σ2 are stability boundary parameters, which are designed based on engineering experience. The parameter range value can be referred to as: σ1=e -1.5 ~e -0.5 ,σ2=e -2.5 ~e -1.5 .
[0025] The step S3 comprises the following steps:
[0026] S31, selecting the rudder deflection command timing signal output by the nominal controller as the input of the network;
[0027] S32. Extract features from the data of the time series signal by discrete Fourier transform, collect time domain feature data, perform DFT (discrete Fourier transform) on the time domain data according to formula (3), extract the frequency domain features of the sample data, and obtain the frequency distribution of the sample data:
[0028]
[0029] Where: N is the length of the observed time series data, x(n) is the time domain sequence of the observed data, is the average value of the time domain data, X(k) is the complex sequence after discrete Fourier transform, and Y(k) represents the magnitude of X(k). Based on the operating frequency range of the missile stabilization and control system, the energy of appropriate frequency points is selected as the characteristic data.
[0030] The step S4 comprises the following steps:
[0031] S41, using the stability level classification obtained in step S3 as network output and the rudder command timing signal data as network input to learn missile state characteristics in different stable states;
[0032] S42, using training data under different stable states, obtaining a missile stable state perception neural network;
[0033] The missile stable state perception neural network includes a one-dimensional convolution layer, a pooling layer, a long short-term memory network layer, a fully connected layer, and a softmax layer;
[0034] One-dimensional convolutional layer is used for simple feature extraction of input data;
[0035] The long short-term memory network layer is used to process the correlation of sequence signals;
[0036] The fully connected layer and the Softmax layer are used to handle the stable state classification.
[0037] The step S5 comprises the following steps:
[0038] S51, based on the original classic controller, a steady-state perception network is added;
[0039] S52, in real time, judging the unstable state of the current missile according to the input rudder command time series data, and then taking the corresponding adaptive adjustment coefficient; the output of the network designed in step S4 has a total of 5 gears, which represent the stable state of the system respectively. Different stable states correspond to different characteristic coefficients. The corresponding relationship is shown in formula (4):
[0040]
[0041] Assume that the original controller's command output is u, and the output after intelligent adaptive adjustment is u * =ku, when the system is in a stable state, k = 1, the function of the regulation law is to make the system tend to be stable, when an unstable state occurs, an instantaneous regulation amount is generated, and then k tends to 1, so the regulation coefficient adaptive law is designed as:
[0042]
[0043] Where: k is the adjustment coefficient, the initial value k0=1, in order to prevent the system from having too extreme adjustments, k has a certain range, k∈[k min ,k max ], which is the maximum adjustment range of the loop gain, k a is the adaptive regulation law, and its initial value is k a0 =0.4, σ is the variable that characterizes the amplitude margin, and is the characteristic coefficient of the output value of the characteristic variable of the Long Short-Term Memory (LSTM) neural network. k2 represents the negative amplitude margin adaptive adjustment gain, k1 represents the positive amplitude margin adaptive adjustment gain, which can be adaptively selected according to the characteristics of the aircraft and the simulation situation, and k3 represents the regression gain.
[0044] S53. Design a steady-state perception network regulation rule, and perform online adaptive adjustment of control parameters based on the output of the perception network to improve the robustness of the controller.
[0045] Compared with the prior art, the present invention has the following beneficial effects:
[0046] 1. The present invention realizes the real-time stability perception of missile flight through the long short-term memory network neural network, which makes up for the deficiency that the traditional missile cannot judge the stability of the missile during flight.
[0047] 2. The present invention designs an online intelligent control parameter adjustment law for missiles, which solves the defect that the parameters of traditional classical controllers cannot be adjusted online.
[0048] 3. The present invention achieves the expansion of the system margin boundary when the preset controller stability margin is insufficient, improves the robustness of the controller, and ensures stable control of the system. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0050] Figure 1 This is the source flow structure diagram of the entire process of the present invention;
[0051] Figure 2 This is a principle block diagram of a controller according to an embodiment of the present invention;
[0052] Figure 3 This is the iterative training process of the steady-state perception network according to an embodiment of the present invention;
[0053] Figure 4 This is a diagram of the overload tracking situation of the pull-bias simulation of the present invention;
[0054] Figure 5 It is the eigenvalue output of the online adaptive controller of the present invention. DETAILED DESCRIPTION
[0055] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be noted that the drawings are all in very simplified form and use non-precise ratios, and are only used for the purpose of conveniently and clearly assisting in explaining the embodiments of the present invention.
[0056] Figure 1 The full-process source flow structure diagram of the present invention introduces an online intelligent adaptive parameter adjustment method for a missile stabilization controller, which includes the following steps:
[0057] S1. Construct a numerical simulation model of the missile's nonlinear dynamics model; deflect important parameters that affect system stability in different combinations to obtain the state output of the missile's dynamics system under different deflection combinations;
[0058] S2. Construct training data labels by quantifying the stable state of the missile stabilization and control system;
[0059] S3, extracting frequency domain features from the training input data;
[0060] S4. Map the missile system's stability to the rudder command timing signal and construct a neural network for steady-state perception.
[0061] S5. Design a controller parameter adaptive adjustment strategy.
[0062] In the step S1:
[0063] Important parameters that affect system stability include rudder efficiency, center of mass, dynamic coefficient, and dynamic pressure;
[0064] In different pull-bias combinations, the system output includes stable output, unstable output, and output divergence;
[0065] By expanding the bias combination, the sample diversity of the training data can be enriched.
[0066] The step S2 comprises the following steps:
[0067] S21. At each observation data point, the amplitude stability margin of the system at that moment is calculated according to the linear system stability method. The comprehensive stability state of the entire observation data is obtained by the time weighting method. This is used as the system stability state metric at the current moment. The calculation is expressed as follows:
[0068]
[0069] Where △ represents the system stability at the current moment, N represents the length of the observation segment data, k represents the number of simulation steps from the start of simulation to the current moment, and k ≥ N, γ represents the attenuation coefficient, γ = 0.995, PGM represents the positive amplitude stability margin, and NGM represents the negative amplitude stability margin.
[0070] S22, the system stable state is represented by a value between [-1, 1];
[0071] S23, classifying the missile stability state into different stability levels according to different instability levels, and obtaining time domain state characteristic data under different instability conditions;
[0072] The stability level classification includes 5 levels, specifically divided into I negative instability, II insufficient negative stability margin, III stable, IV insufficient positive stability margin, and V positive instability. The specific classification standard is shown in formula (2):
[0073]
[0074] Among them: σ1, σ2 are stability boundary parameters, which are designed based on engineering experience. The parameter range value can be referred to as: σ1=e -1.5 ~e -0.5 ,σ2=e -2.5 ~e -1.5 .
[0075] The step S3 comprises the following steps:
[0076] S31, selecting the rudder deflection command timing signal output by the nominal controller as the input of the network;
[0077] S32. Extract features from the data of the time series signal by discrete Fourier transform, collect time domain feature data, perform DFT (discrete Fourier transform) on the time domain data according to formula (3), extract the frequency domain features of the sample data, and obtain the frequency distribution of the sample data:
[0078]
[0079] Where: N is the length of the observed time series data, x(n) is the time domain sequence of the observed data, is the average value of the time domain data, X(k) is the complex sequence after discrete Fourier transform, and Y(k) represents the magnitude of X(k). Based on the operating frequency range of the missile stabilization and control system, the energy of appropriate frequency points is selected as the characteristic data.
[0080] The step S4 comprises the following steps:
[0081] S41, using the stability level classification obtained in step S3 as network output and the rudder command timing signal data as network input to learn missile state characteristics in different stable states;
[0082] S42, using training data under different stable states, obtaining a missile stable state perception neural network;
[0083] The missile stable state perception neural network includes a one-dimensional convolution layer, a pooling layer, a long short-term memory network layer, a fully connected layer, and a softmax layer;
[0084] One-dimensional convolutional layer is used for simple feature extraction of input data;
[0085] The long short-term memory network layer is used to process the correlation of sequence signals;
[0086] The fully connected layer and the Softmax layer are used to handle the stable state classification.
[0087] The step S5 comprises the following steps:
[0088] S51, based on the original classic controller, a steady-state perception network is added;
[0089] S52, in real time, judging the unstable state of the current missile according to the input rudder command time series data, and then taking the corresponding adaptive adjustment coefficient; the output of the network designed in step S4 has a total of 5 gears, which represent the stable state of the system respectively. Different stable states correspond to different characteristic coefficients. The corresponding relationship is shown in formula (4):
[0090]
[0091] Figure 2 This is a block diagram of the controller principle of an embodiment of the present invention. Assume that the original controller's command output is u, and the output after intelligent adaptive adjustment is u * =ku, when the system is in a stable state, k = 1, the function of the regulation law is to make the system tend to be stable, when an unstable state occurs, an instantaneous regulation amount is generated, and then k tends to 1, so the regulation coefficient adaptive law is designed as:
[0092]
[0093] Where: k is the adjustment coefficient, the initial value k0=1, in order to prevent the system from having too extreme adjustments, k has a certain range, k∈[k min ,k max ], which is the maximum adjustment range of the loop gain, k a is the adaptive regulation law, and its initial value is k a0 =0.4, σ is the variable that characterizes the amplitude margin, and is the characteristic coefficient of the output value of the characteristic variable of the Long Short-Term Memory (LSTM) neural network. k2 represents the negative amplitude margin adaptive adjustment gain, k1 represents the positive amplitude margin adaptive adjustment gain, which can be adaptively selected according to the characteristics of the aircraft and the simulation situation, and k3 represents the regression gain.
[0094] S53. Design a steady-state perception network regulation rule, and perform online adaptive adjustment of control parameters based on the output of the perception network to improve the robustness of the controller.
[0095] Steady-state perception network training
[0096] Table 1 shows the simulation deviation conditions of an embodiment of the present invention. According to the simulation deviation conditions described in Table 1, 100 groups of combined deviation simulations are performed on a missile to generate an original training data set. The observation data window length is 1024, the sampling time is 2.5ms, and a total of 42150 groups of data are generated. The data is divided into 70% for training and 30% for testing. Finally, 33720 groups of data are used as training data and 8430 groups of data are used as testing data. The training platform uses matlab 2021b, the system is Inetl(R)Core(TM) i5-9600KF CPU@3.7GHz, and the memory is 16G. The training accuracy and loss iteration diagram are as follows: Figure 3 As shown, the overall training accuracy is 93.58%.
[0097] Table 1 Simulation pulling conditions of the embodiment of the present invention
[0098]
[0099] Closed-loop simulation verification
[0100] The trained steady-state perception network was incorporated into a six-degree-of-freedom model for closed-loop simulation to verify the feasibility of the controller. Simulation conditions: Under nominal data, the steering efficiency was reduced by 30%, and the intelligent adaptive controller parameters were: k1 = 5, k2 = 10, and k3 = 0.1.
[0101] Figure 4 The simulation results show that under the nominal controller, the system diverges, that is, even under the nominal design with a certain stability margin, it is still impossible to achieve stable control of the system when encountering a large range of rudder effect deviation. After adding the controller of the adaptive intelligent parameter adjustment method of the present invention, the results of the controller parameter adaptive adjustment are shown in Figure 2. Figure 5 The intelligent controller sensed the unstable state of the system and made timely online adjustments to the control gain, ultimately achieving stable control. This shows that when the preset stability margin is insufficient, the inventive method can achieve rapid online correction of controller parameters, thereby ensuring the stability of the system.
[0102] Although the present invention has been disclosed above with reference to preferred embodiments, this is not intended to limit the present invention. Any person skilled in the art may utilize the methods and techniques disclosed above to make possible changes and modifications to the technical solutions of the present invention without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent variations, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the scope of protection of the technical solutions of the present invention. Matters not described in detail in the present specification are common knowledge to those skilled in the art.
Claims
1. A method for online intelligent adaptive parameter adjustment of a missile stabilization controller, characterized in that: The following steps are involved: S1. Construct a numerical simulation model of the missile's nonlinear dynamics model; deflect important parameters that affect system stability in different combinations to obtain the state output of the missile's dynamics system under different deflection combinations; S2. Construct training data labels by quantifying the stable state of the missile stabilization and control system; S3, extracting frequency domain features from the training input data; S4. Map the missile system's stability to the rudder command timing signal and construct a neural network for steady-state perception. S5. Design a controller parameter adaptive adjustment strategy.
2. The method for online intelligent adaptive parameter adjustment of a missile stabilization controller according to claim 1, characterized in that: In the step S1: Important parameters that affect system stability include rudder effectiveness, center of mass, dynamic coefficient, and dynamic pressure.
3. The online intelligent adaptive parameter adjustment method for a missile stabilization controller according to claim 1 is characterized in that: The step S2 comprises the following steps: S21. At each observation data point, the amplitude stability margin of the system at that moment is calculated using the linear system stability method. The comprehensive stability state of the entire observation data segment is obtained using the time-weighted method. This comprehensive stability state is used as the system stability state metric at the current moment. S22, the system stable state is represented by a value between [-1, 1]; S23. Classify the missile's stability level according to different instability levels, and obtain time domain state characteristic data under different instability conditions.
4. The method for online intelligent adaptive parameter adjustment of a missile stabilization controller according to claim 3, characterized in that: In step S23, the stability level classification includes 5 levels, specifically divided into I negative instability, II insufficient negative stability margin, III stable, IV insufficient positive stability margin, and V positive instability.
5. The online intelligent adaptive parameter adjustment method for a missile stabilization controller according to claim 1 is characterized in that: The step S3 comprises the following steps: S31, selecting the rudder deflection command timing signal output by the nominal controller as the input of the network; S32 , extracting features from the data of the time series signal, collecting time domain feature data, performing discrete Fourier transform on the time domain data, extracting frequency domain features of the sample data, and obtaining a frequency distribution of the sample data.
6. The method for online intelligent adaptive parameter adjustment of a missile stabilization controller according to claim 5, characterized in that: In step S32, the length of the collected time domain feature data is the same as the length of the observed time series data.
7. The method for online intelligent adaptive parameter adjustment of a missile stabilization controller according to claim 5, characterized in that: In step S32, the method for extracting features from the time series signal data is discrete Fourier transform.
8. The method for online intelligent adaptive parameter adjustment of a missile stabilization controller according to claim 1, characterized in that: The step S4 comprises the following steps: S41, using the stability level classification obtained in step S3 as network output and the rudder command timing signal data as network input to learn missile state characteristics in different stable states; S42. Utilize the training data under different stable states to obtain a missile stable state perception neural network.
9. The method for online intelligent adaptive parameter adjustment of a missile stabilization controller according to claim 7, characterized in that: The missile stable state perception neural network in step S42 includes a one-dimensional convolutional layer, a pooling layer, a long short-term memory network layer, a fully connected layer, and a softmax layer; One-dimensional convolutional layer is used for simple feature extraction of input data; The long short-term memory network layer is used to process the correlation of sequence signals; The fully connected layer and the Softmax layer are used to handle the stable state classification.
10. The method for online intelligent adaptive parameter adjustment of a missile stabilization controller according to claim 1, characterized in that: The step S5 comprises the following steps: S51, based on the original classic controller, a steady-state perception network is added; S52, judging the current instability state of the missile in real time based on the input rudder command time series data, and then adopting a corresponding adaptive adjustment coefficient; S53. Design a steady-state perception network regulation rule, and perform online adaptive adjustment of control parameters based on the output of the perception network to improve the robustness of the controller.