Unmanned aerial vehicle interference observer design method based on RBF neural network
By designing an interference observer based on RBF neural networks, the problems of high accuracy and stability in interference estimation for UAVs in complex environments are solved, achieving fast and accurate interference compensation and improving the robustness and adaptability of UAVs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-22
- Publication Date
- 2026-04-07
AI Technical Summary
Existing UAV interference estimation methods struggle to achieve high-precision and stable interference compensation in complex dynamic environments. In particular, offline-trained neural networks have poor generalization capabilities, nonlinear observers rely on the drift of dynamic model parameters, and extended state observers are prone to flutter and noise amplification.
An interference observer design method based on RBF neural network is adopted. By unifying the description of UAV motion and interference information, a nonlinear interference observer is constructed and combined with an RBF neural network approximator to achieve fast and accurate estimation of lumped interference.
It improves the robustness and adaptability of UAVs in complex environments, reduces system complexity and engineering implementation difficulty, achieves shorter convergence time and higher estimation accuracy, and provides real-time and reliable interference compensation support.
Smart Images
Figure CN121809241A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of unmanned aerial vehicle interference estimation, and particularly relates to a design method of an unmanned aerial vehicle interference observer based on a RBF neural network. BACKGROUND
[0002] With the wide application of unmanned aerial vehicles in aerial photography, power inspection and emergency rescue, the operation environment of the unmanned aerial vehicles gradually extends from open airspace to complex electromagnetic environment and airflow disturbance area. In actual flight, the unmanned aerial vehicles are easily affected by wind, electromagnetic interference and load fluctuation, which directly damages the attitude stability and trajectory tracking accuracy of the unmanned aerial vehicles, and even leads to flight out of control. Therefore, how to estimate the interference suffered by the unmanned aerial vehicles in real time and accurately and provide reliable interference compensation basis for flight control becomes a core technical problem for ensuring the safety and control performance of the unmanned aerial vehicles in complex environments.
[0003] At present, the technical means for unmanned aerial vehicle interference estimation mainly includes three types. The first type is a neural network approximation method based on offline training. The method constructs a sample database containing multiple typical interference scenarios, uses the nonlinear fitting capability of a neural network (such as a BP neural network and a CNN), and offline trains a mapping model between interference and unmanned aerial vehicle state parameters (such as angular velocity, acceleration and attitude angle). In actual application, the pre-trained model is called to realize interference estimation by inputting real-time state parameters. The second type is a nonlinear interference observer. The nonlinear interference observer is based on the accurate dynamics model of the unmanned aerial vehicle, treats the interference as unknown input by designing a nonlinear observation function, and inversely estimates the interference information in real time by using the deviation between the measured value and the estimated value of the state variable. The nonlinear interference observer is suitable for scenarios where the model structure is known but the parameters are perturbed. The third type is an extended state observer. The extended state observer does not need to rely on an accurate model, treats the interference and unmodeled dynamics as a lumped disturbance and expands it into a new state variable, and realizes the simultaneous estimation of the state variable and the total disturbance by a high-gain observation algorithm. The extended state observer is widely used in scenarios with strong model uncertainty.
[0004] However, the existing interference estimation methods still have obvious shortcomings, which are difficult to meet the high-precision control requirements of unmanned aerial vehicles in complex dynamic environments. For the neural network approximation method of offline training, the estimation performance is severely dependent on the integrity and representativeness of the sample database. However, the actual interference scene has diversity, randomness and time-varying characteristics. It is not only costly and time-consuming to build a sample database covering all possible scenarios, but also the estimation accuracy will be greatly reduced and the generalization ability will be poor when facing new types of interference not included in the training set. For the nonlinear disturbance observer, the design premise is to accurately master the dynamic model parameters (such as mass, moment of inertia, and aerodynamic coefficient) of the unmanned aerial vehicle. However, in actual flight, the model parameters will dynamically drift due to changes in load, equipment wear, and environmental temperature changes, resulting in a mismatch between the observer and the actual model, and thus a large estimation error. For the extended state observer, although it has low dependence on the model, in order to track the rapidly changing disturbance, a high observation gain is usually required, which will amplify the influence of measurement noise, resulting in a sharp fluctuation of the disturbance estimation value. It is also difficult to ensure the estimation accuracy when the disturbance changes gently, and it is easy to cause chattering phenomenon, affecting the flight stability of the unmanned aerial vehicle.
[0005] In summary, a new disturbance observation method needs to be developed to overcome the shortcomings of existing observation methods and achieve rapid and accurate estimation of unmanned aerial vehicle disturbances, providing a solid foundation for unmanned aerial vehicle disturbance rejection control. SUMMARY
[0006] The present application aims to provide a method for designing an unmanned aerial vehicle disturbance observer based on RBF neural networks to achieve accurate disturbance estimation.
[0007] To achieve the purpose of the present application, the present application provides a method for designing an unmanned aerial vehicle disturbance observer based on RBF neural networks, comprising the following steps:
[0008] Step 1, collect the motion information, control input information and lumped disturbance information of the unmanned aerial vehicle;
[0009] Step 2, unify the velocity sub-model and angular velocity sub-model based on the motion information, control input information and lumped disturbance information of the unmanned aerial vehicle, and obtain the disturbance estimation model of the unmanned aerial vehicle;
[0010] Step 3, according to the disturbance estimation model, combine the motion information, control input information and lumped disturbance information to construct a nonlinear disturbance observer to preliminarily estimate the lumped disturbance, and obtain the preliminary estimation value of the lumped disturbance;
[0011] Step 4, use the preliminary estimation value of the lumped disturbance to obtain the estimation error and its approximation value of the nonlinear disturbance observer through the neural network approximator constructed by RBF;
[0012] Step 5, according to the nonlinear disturbance observer and the neural network approximator, combining the preliminary estimate value of the lumped disturbance and the approximation value of the nonlinear disturbance observer estimation error, constructing the RBF neural network disturbance observer, and obtaining the final estimate value of the lumped disturbance.
[0013] A computer program product comprises computer program instructions, which, when executed on a computer, cause the computer to perform the above-mentioned UAV disturbance observer design method.
[0014] Compared with the prior art, the significant progress of the present application is that (1) the present application avoids the cumbersome sample library establishment process, reduces the system complexity and engineering implementation difficulty; (2) the present application realizes shorter convergence time and higher estimation accuracy through the powerful nonlinear fitting capability of the RBF neural network; this provides real-time and reliable key support for the overall compensation of the control system to the disturbance effect, thereby greatly improving the robustness and adaptability of the UAV in the complex unknown environment.
[0015] To more clearly illustrate the functional characteristics and structural parameters of the present application, the following further describes the present application in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS
[0016] The accompanying drawings, which are included to provide a further understanding of the present application, constitute a part of this application, illustrate embodiments of the present application and specific embodiments thereof, and are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings:
[0017] Figure 1 is a step flow chart of the present application;
[0018] Figure 2 is an RBF neural network structure diagram of the present application;
[0019] Figure 3 is a UAV acceleration x-direction disturbance estimation curve of the embodiment of the present application;
[0020] Figure 4 is a UAV acceleration y-direction disturbance estimation curve of the embodiment of the present application;
[0021] Figure 5 is a UAV acceleration z-direction disturbance estimation curve of the embodiment of the present application. DETAILED DESCRIPTION
[0022] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of the present application.
[0023] The method comprises the following steps: Figure 1 , comprising the following steps:
[0024] Step 1, collecting motion information, control input information and lumped interference information of the unmanned aerial vehicle;
[0025] Step 2, uniformly describing a speed sub-model and an angular velocity sub-model according to the motion information, the control input information and the lumped interference information of the unmanned aerial vehicle, and obtaining an interference estimation model of the unmanned aerial vehicle;
[0026] Step 3, constructing a nonlinear interference observer to preliminarily estimate the lumped interference according to the interference estimation model and in combination with the motion information, the control input information and the lumped interference information, and obtaining a preliminary estimation value of the lumped interference;
[0027] Step 4, obtaining an estimation error of the nonlinear interference observer and an approximation value thereof by using the preliminary estimation value of the lumped interference and a neural network approximator constructed by RBF;
[0028] Step 5, combining the preliminary estimation value of the lumped interference and the approximation value of the estimation error of the nonlinear interference observer according to the nonlinear interference observer and the neural network approximator, constructing an RBF neural network interference observer, and obtaining a final estimation value of the lumped interference.
[0029] The motion information of step 1 comprises speed and attitude angular velocity, and specifically, state measurement values are obtained by using an accelerometer, a rate gyro and a navigation algorithm; the control input information comprises unmanned aerial vehicle acceleration and propeller speed; and the lumped interference information comprises external wind interference and load fluctuation.
[0030] The step 2 is specifically shown in the following formula:
[0031] ;
[0032] wherein, is an unmanned aerial vehicle motion state, specifically, unmanned aerial vehicle speed or attitude angular velocity; is composed of motion information; is a control input weight matrix; is control input information; This refers to the lumped interference information received by the drone.
[0033] Step 3 is specifically shown in the following formula:
[0034] ;
[0035] in, For centralized interference Preliminary estimate; For the intermediate variables of the interference observer; ; The vector represents the nonlinear function to be designed.
[0036] Combination Figure 2 Step 4 includes the following steps:
[0037] Step 4-1: Based on the preliminary estimate of the lumped interference, obtain the RBF neural network expression of the nonlinear interference observer estimation error through the neural network constructed by RBF;
[0038] Step 4-2: Based on the RBF neural network expression of the nonlinear disturbance observer estimation error described above, design the neural network update law to obtain the approximate value of the nonlinear disturbance observer estimation error.
[0039] Step 4-3: Based on the above neural network update law and the approximation value of the nonlinear disturbance observer estimation error, derive the RBF neural network gradient, and then obtain the neural network parameter iteration formula.
[0040] The RBF neural network in step 4-1 is specifically shown in the following formula:
[0041] ;
[0042] The estimation error of the nonlinear disturbance observer is: , The ideal neural network weight matrix; The vector composed of RBFs is shown in the following formula:
[0043] ;
[0044] in, ; For the hidden layer The center vector of each node; is the base width of the hidden layer node.
[0045] The neural network update law and the estimation error approximation value in step 4-2 are specifically shown in the following formulas:
[0046] ;
[0047] in, for Approximation value; The learning rate for the neural network; , for The estimated value is shown in the following formula:
[0048] ;
[0049] in, For interference The final estimated value. The RBF neural network structure is as follows: Figure 2 As shown. The network input vector is The output vector is ; Network weight vector matrix The Middle line, number The elements of the column.
[0050] The RBF neural network gradient in step 4-3 is specifically shown in the following formula:
[0051] ;
[0052] in, , , These are the parameters of the neural network. , , The gradient;
[0053] The iterative formula for updating the neural network parameters using gradient descent is:
[0054] ;
[0055] in, The learning rate for the neural network; This is the momentum term factor.
[0056] The interference observer in step 5 is specifically shown in the following formula:
[0057] .
[0058] Example
[0059] Taking the UAV velocity subsystem as an example, the interference observer improved by the RBF neural network is applied to the estimation of interference acceleration to verify the effectiveness and advantages of the algorithm.
[0060] The motion model of the drone's center of mass is as follows:
[0061] ;
[0062] in, Represents the three-dimensional velocity vector of the UAV; The acceleration generated after the drone adjusts its attitude can be regarded as a control input; The acceleration vector generated by gravity, wind interference, etc., on the drone represents system interference.
[0063] The system interference is set as follows:
[0064] ;
[0065] The interference observer is designed as follows:
[0066] ;
[0067] The relevant parameter values are:
[0068] .
[0069] like Figures 3-5 As shown, traditional nonlinear observers suffer from significant periodicity estimation errors when estimating periodic signals. In contrast, the novel observer proposed in this invention utilizes the nonlinear approximation and online learning capabilities of RBFNN to quickly approximate and compensate for the aforementioned estimation errors, enabling it to achieve both faster convergence speed and higher accuracy in estimating time-varying signals.
[0070] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A method for designing a UAV interference observer based on an RBF neural network, characterized in that, Includes the following steps: Step 1: Collect the UAV's motion information, control input information, and lumped interference information; Step 2: Based on the UAV's motion information, control input information, and lumped interference information, unify the velocity sub-model and angular velocity sub-model to obtain the UAV's interference estimation model; Step 3: Based on the interference estimation model, and combined with the motion information, control input information, and lumped interference information, construct a nonlinear interference observer to perform a preliminary estimation of the lumped interference and obtain a preliminary estimate of the lumped interference. Step 4: Using the preliminary estimate of the lumped interference, obtain the estimation error of the nonlinear interference observer and its approximation value through the neural network approximator constructed by RBF; Step 5: Based on the nonlinear interference observer and the neural network approximator, combine the preliminary estimate of the lumped interference with the approximation value of the estimation error of the nonlinear interference observer to construct the RBF neural network interference observer and obtain the final estimate of the lumped interference.
2. The method for designing a UAV jamming observer based on an RBF neural network according to claim 1, characterized in that, The motion information in step 1 includes velocity and attitude angular velocity. Specifically, it is obtained by using accelerometers and rate gyroscopes in combination with navigation algorithms to obtain state measurement values. The control input information includes UAV acceleration and propeller speed. The lumped interference information includes external wind interference and load fluctuations.
3. The method for designing a UAV jamming observer based on an RBF neural network according to claim 2, characterized in that, Step 2 is specifically shown in the following formula: ； in, This refers to the drone's motion state, specifically its speed or attitude angular velocity; Composed of motion information; To control the input weight matrix; To control input information; This refers to the lumped interference information received by the drone.
4. The method for designing a UAV jamming observer based on an RBF neural network according to claim 3, characterized in that, Step 3 is specifically shown in the following formula: ； in, For centralized interference Preliminary estimate; For the intermediate variables of the interference observer; ; The vector represents the nonlinear function to be designed.
5. The method for designing a UAV jamming observer based on an RBF neural network according to claim 4, characterized in that, Step 4 includes the following steps: Step 4-1: Based on the preliminary estimate of the lumped interference, obtain the RBF neural network expression of the nonlinear interference observer estimation error through the neural network constructed by RBF; Step 4-2: Based on the RBF neural network expression of the nonlinear disturbance observer estimation error described above, design the neural network update law to obtain the approximate value of the nonlinear disturbance observer estimation error. Step 4-3: Based on the above neural network update law and the approximation value of the nonlinear disturbance observer estimation error, derive the RBF neural network gradient, and then obtain the neural network parameter iteration formula.
6. The method for designing a UAV jamming observer based on an RBF neural network according to claim 5, characterized in that, The RBF neural network in step 4-1 is specifically shown in the following formula: ； The estimation error of the nonlinear disturbance observer is: , The ideal neural network weight matrix; The vector composed of RBFs is shown in the following formula: ； in, ; For the hidden layer The center vector of each node; is the base width of the hidden layer node.
7. The method for designing a UAV jamming observer based on an RBF neural network according to claim 6, characterized in that, The neural network update law and the estimation error approximation value in step 4-2 are specifically shown in the following formulas: ； in, for Approximation value; The learning rate for the neural network; , for The estimated value is shown in the following formula: ； in, For interference The final estimated value. The RBF neural network structure is shown in Figure 2. The network input vector is... The output vector is ; Network weight vector matrix The Middle line, number The elements of the column.
8. The method for designing a UAV jamming observer based on an RBF neural network according to claim 7, characterized in that, The RBF neural network gradient in step 4-3 is specifically shown in the following formula: ； in, , , Neural network parameters , , The gradient; The iterative formula for updating the neural network parameters using gradient descent is: ； in, The learning rate for the neural network; This is the momentum term factor.
9. The method for designing a UAV jamming observer based on an RBF neural network according to claim 8, characterized in that, The interference observer in step 5 is specifically shown in the following formula: 。 10. A computer program product comprising computer program instructions, characterized in that, When the computer program instructions are executed on a computer, the computer causes the computer to perform the method as described in any one of claims 1 to 9.