CN-121979249-A - Finite time cross-domain aircraft cluster formation control method based on neural network

CN121979249ACN 121979249 ACN121979249 ACN 121979249ACN-121979249-A

Abstract

The invention provides a finite time cross-domain aircraft cluster formation control method based on a neural network, which relates to the field of aircraft control and specifically comprises the steps of establishing a unified nonlinear non-strict feedback system aiming at dynamics characteristics of a cross-domain aircraft in different domains, collecting real-time flight states and position information of a plurality of followers, designing an improved Tan type nonlinear mapping function, identifying an initial output constraint type of the system to obtain unconstrained new variables, constructing a dual neural network combining a shared network and a proprietary network, designing a reverse-footwork combined switching function and a nonsingular fast finite time control strategy to control structural output of the dual neural network, and designing a switching threshold event triggering mechanism to enable the unmanned aircraft cluster to have different responses in different environments. The technical scheme of the invention solves the problem that the formation control method in the prior art does not consider mode switching, dynamic coupling and cooperative consistency in cross-domain cooperation.

Inventors

LIU ZENGKAI
Kou Hongyuan
CHEN YUNSAI
ZHANG DONG
JIANG QINGHUA
GAO YONG

Assignees

青岛哈尔滨工程大学创新发展中心
哈尔滨工程大学

Dates

Publication Date: 20260505
Application Date: 20260330

Claims (10)

1. The finite time cross-domain aircraft cluster formation control method based on the neural network is characterized by comprising the following steps of: s1, establishing a unified nonlinear non-strict feedback system aiming at dynamics characteristics of a cross-domain aircraft in different domains; s2, acquiring real-time flight states and position information of a plurality of followers, and sending the states of the leaders to the plurality of followers, wherein the plurality of followers are communicated with each other; s3, designing an improved Tan type nonlinear mapping function, analyzing the original state received by a sensor, and identifying the initial output constraint type of the system to obtain an unconstrained new variable; S4, constructing a dual neural network combining a shared network and an exclusive network, wherein the shared network learns general dynamics characteristics of the cross-domain aircraft, adopts the same network structure and initial weight, combines local gradient and distributed consistency items, learns cross-domain commonality; s5, designing a backstepping method, combining a switching function and a nonsingular fast finite time control strategy, and controlling the output of the dual neural network structure; S6, designing a trigger mechanism of a switching threshold event, combining trigger conditions of combination of a fixed threshold and a relative threshold, and enabling the unmanned aerial vehicle group to have different responses in different environments based on double judgment of a weight change rate and a time interval.
2. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 1, wherein the step S1 specifically comprises the following steps: S1.1, establishing a heterogeneous cross-domain aircraft cluster consisting of 1 leader and N followers, wherein a dynamics model of each follower is described by a nonlinear non-strict feedback system: ; Track of leader : A bounded smooth function is known; Wherein, the Is the first The state variables of the individual follower aircraft, For unknown smooth nonlinear functions, including aerodynamic, hydrodynamic and environmental disturbance factors, Is the first The system order of the individual follower aircraft, Is the first A control input of the individual follower aircraft, Is the first Highest order in a state vector of a follower aircraft Is used for the control of the state of (a), Is the first An output of the individual follower aircraft; s1.2, constraining the track of the leader and the output of the follower: And Is the derivative of (2) Are known and bounded; Output of the system Is constrained, i.e Or (b) , wherein, , ; Is that Is set in accordance with the constant constraint of (a), Is that Is a time-varying constraint of (a), Is a constant value, and is used for the treatment of the skin, As a time-varying function.
3. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 1, wherein the step S3 specifically comprises the following steps: s3.1, when the output constraint type is constant output, mapping the function The method comprises the following steps: ; Wherein, the As a result of the original output variable being constrained, Is a tangent function; S3.2, when the output constraint type is time-varying output, mapping function The method comprises the following steps: ; s3.3, order Ensuring that no singular point appears in the system.
4. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 1, wherein the step S4 specifically comprises the following steps: s4.1, dedicated network Compensation The specific dynamics of the individual follower aircraft, the proprietary network output is: ; Wherein, the Is the first The status of the individual follower aircraft is determined, Is the transpose of the real-time estimated weights of the proprietary neural network, Is a proprietary network RBF basis function, Is the state vector of the exclusive network system, Is an unknown dynamics estimate of each individual of the proprietary network; s4.2, the sharing network learns general dynamics common to all follower aircrafts, and the sharing network outputs are: ; Wherein, the Is a transpose of the network weights of the shared neural network, Is a shared network RBF basis function, Is an unknown kinetic estimate of each individual sharing the network, Is a shared network system state vector; S4.3, th The total unknown nonlinear approximation of the individual follower aircraft is: ; Wherein, the And Real-time weight estimation of shared and exclusive networks respectively; S4.4, judging the network performance according to the error, if the shared network error is smaller than the exclusive network error, namely The shared network performance is better, otherwise: the exclusive network is more suitable for the current state, and the fusion weight is dynamically adjusted Dual neural network structure output The method comprises the following steps: 。
5. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 4, wherein the weight estimation of the private neural network in step S4.1 specifically comprises the following steps: s4.1.1, for a true unknown kinetic function, there is an ideal optimal weight Such that: ; Wherein, the As a function of the non-linearity, For the optimal weight value to be the optimal weight value, Is that Is to be used in the present invention, As a function of the basis vector, Is an approximation residual; Estimation output of proprietary neural network The method comprises the following steps: ; Wherein, the The weight is estimated; s4.1.2 defining approximation error of network estimated value and true value The method comprises the following steps: ; Defining weight error The difference between the true weight and the estimated weight is: ; substitution expansion is carried out to obtain an approximation error equation: ; s4.1.3, to ensure the stability of the subsequent closed loop system, deriving a weight update law of a continuous form of a proprietary network by combining with Lyapunov stability analysis of the whole system The method comprises the following steps: ; Wherein, the In order to learn the rate matrix of the device, For systematic tracking errors defined in the subsequent back-stepping, Is the attenuation coefficient of the exclusive network and ; Discretizing the continuous update law of the weight of the proprietary network to directly obtain the final discrete update process of the weight of the proprietary network in the actual engineering application: ; Wherein, the And Dedicated network weight estimation for the next discrete time and the current time respectively; a learning rate matrix for the exclusive network; Is the first The following aircraft is at the first Step back tracking error.
6. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 4, wherein the weight estimation of the shared neural network in step S4.2 is specifically: ; Wherein, the Is the first The following aircraft is at the first In the step back-step control, the shared network weight value of the next discrete moment is estimated; the shared network weight estimation at the current discrete moment; Is a learning rate matrix of the shared network; Is the RBF basis function vector of the shared network; Is the first The following aircraft is at the first Step back tracking error; Is the first Estimating the shared network weight of each follower aircraft at the current moment; is a communication topology adjacency matrix element; Is the first A set of neighbors of the individual follower aircraft; Is the attenuation coefficient of the shared network, and 。
7. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 1, wherein the step S5 specifically comprises the following steps: S5.1, utilizing the back-step tracking error to approximate the influence of estimated residual errors, and designing a residual error compensation term according to the influence of the estimated residual errors: ; Wherein, the To compensate for gain; Is the first The individual follower aircraft is at the highest order Tracking errors of the system; S5.2, gradually designing a virtual control law for each follower aircraft by adopting a self-adaptive back-stepping method, defining coordinate transformation, and defining a first error variable : ; Wherein, the Is the connection coefficient of the leader and the connection coefficient of the leader, Is a mapping function; Redefinition of the first Error variable : ; Wherein, the Is the first Virtual control laws of step design.
8. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 7, wherein the step S5 further comprises the steps of: S5.3, combining the double neural network approximation error, performing a back-step method virtual control law and overall Lyapunov stability design, aiming at the first Error dynamics of steps combined with system model And the unknown nonlinear function is replaced by a double neural network approximation form Obtaining an error dynamics equation: ; Wherein, the Is that Is used for the purpose of determining the derivative of (c), Is that Is to be used in the present invention, First, the The following aircraft is at the first The difference between the true weight and the estimated weight of the step, Is the first The following aircraft is at the first The step of sharing the network RBF basis functions, First, the The following aircraft is at the first The approximation of the step is made to the residual, Is that Is a derivative of (2); s5.4, in order to ensure the stability of the closed loop system, design the th Lyapunov function of step The system tracking error and the neural network weight estimation error are jointly included in an energy function: ; Wherein, the As the learning rate matrix, the optimal weight is constant, so For a pair of Derivative is obtained by substituting an error dynamics equation to be developed: ; Wherein, the Is that Is a derivative of (2); S5.5, to eliminate unknown weight cross terms Order-making Deducing a theoretical continuous update law of the estimation weight of the private network in the step S4: ; At the same time, in order to make Satisfying negative definite conditions, i.e. Order-making Is virtual control law According to this design The virtual control law of the steps is as follows: ; Wherein, the Is the first Linear gain of steps; Is the first In the step back step derivation, dynamic estimation values output by a dual neural network are utilized; the first-order time derivative of the virtual control law of the previous order is used as a feedforward term for eliminating coupling dynamics in the process of the back-stepping recursion; Is the first The residual error of the step compensates for the gain.
9. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 8, wherein the step S5 further comprises the steps of: s5.6, designing a non-singular switching function to avoid singularity: ; Wherein, the Is an error variable in a back-step design ; Is a handover threshold; Is an exponential parameter, wherein, , Is positive odd number, and ; S5.7, final control law The method integrates back-stepping basic control, nonsingular rapid finite time control and dual neural network compensation, and is designed as follows: ; Wherein, the In order to track the error in the tracking, In order to control the gain of the gain control, Is the first The derivative of the step virtual control law, And outputting an estimated value for the dual neural network structure.
10. The method for controlling formation of a finite time cross-domain aircraft cluster based on a neural network according to claim 1, wherein the step S6 specifically comprises the following steps: S6.1, definition of Individual follower aircraft on Measurement error of time of day The method comprises the following steps: ; Wherein, the Ideal control input for the system to execute at the current moment; Is the first The moment when the follower aircraft successfully triggers the event and updates the state last time; a control instruction for the actuator currently being actually held for execution; S6.2, designing a bimodal switching threshold triggering condition when , wherein, For a preset control input boundary constant, a relative threshold strategy is adopted, and the next event triggering moment The determination conditions of (2) are: ; Wherein, the A minimum time infinitesimal that represents a condition being met; Is a relative threshold coefficient; is extremely small positive constant; When (when) The trigger determination condition is switched to: ; Wherein, the A fixed safety threshold value is set; S6.3, when the system state meets any trigger condition in the step S6.2, the current moment is marked as At this time, the first The individual follower aircraft will broadcast the latest status information to the neighbors and will be up to date Giving the control update to the executor, if the condition is not satisfied, maintaining the execution of the instruction at the previous time 。

Description

Finite time cross-domain aircraft cluster formation control method based on neural network Technical Field The invention relates to the field of aircraft control, in particular to a finite time cross-domain aircraft cluster formation control method based on a neural network. Background With the wide application of unmanned aerial vehicles, unmanned boats, unmanned vehicles and other unmanned systems in various fields, cross-domain collaborative operation has become an important direction for the development of intelligent unmanned systems in the future. The cross-domain aircraft cluster has cross-medium operation capability between the air and the water surface, and can execute complex tasks such as wide area monitoring, collaborative search and rescue, distributed detection and the like. However, the formation control of the cross-domain aircraft cluster faces serious challenges under the influence of multiple factors such as dynamic switching operation domain, strong nonlinear coupling dynamics, multi-agent communication limitation, external environment disturbance and the like. At a technical level, the dynamics model of a cross-domain aircraft has a high degree of nonlinearity and time-variability. Cross-domain aircraft are aerodynamically dominated in the air domain and hydrodynamically affected in the water domain, with substantial differences in fluid drag, propulsion efficiency and stability. In the cross-domain conversion process, dynamic parameters (such as mass distribution, inertia and damping coefficient) of the aircraft are changed drastically, and a traditional control method (such as PID control and sliding mode control) based on an accurate model is difficult to adapt, so that control performance is reduced and even instability is caused. In addition, the existing formation control method is mostly based on an asymptotic stability theory, so that the system can not be guaranteed to quickly converge in a limited time, and the task execution efficiency is affected. When external disturbance such as stormy waves and water flow exists, the traditional control method lacks an effective online compensation mechanism, and formation form damage or tracking error increase is easy to cause. In terms of communication, most methods assume a fixed topology structure, and are difficult to adapt to dynamic topology changes (such as node faults or link breaks), so that flexibility and reliability of the system are limited. The current state of research at home and abroad shows that some work attempts have been made to introduce neural networks to enhance the self-adaptive capacity of the system. For example, the rotor unmanned aerial vehicle self-adaptive control method based on the switching system processes dynamic change in a single domain through mode switching, but does not relate to cross-domain coordination, and the multi-mode smooth switching control method of the carrier unmanned aerial vehicle improves adaptability, but lacks systematic guarantee on limited time convergence performance. In addition, the prior art aims at single domain formation control, and the problems of mode switching, dynamic coupling, cooperative consistency and the like in cross-domain cooperation are not considered. Therefore, there is a need for a neural network-based finite time cross-domain aircraft cluster formation control method that can accommodate cross-domain dynamics, has strong robustness and rapid convergence capabilities. Disclosure of Invention The invention mainly aims to provide a finite time cross-domain aircraft cluster formation control method based on a neural network, which is used for solving the problem that the formation control method in the prior art does not consider mode switching, dynamic coupling and cooperative consistency in cross-domain cooperation. In order to achieve the above purpose, the invention provides a finite time cross-domain aircraft cluster formation control method based on a neural network, which specifically comprises the following steps: s1, a unified nonlinear non-strict feedback system is established aiming at dynamics characteristics of a cross-domain aircraft in different domains. S2, acquiring real-time flight states and position information of a plurality of followers, and sending the states of the leaders to the plurality of followers, wherein the plurality of followers are communicated with each other. S3, designing an improved Tan type nonlinear mapping function, analyzing the original state received by the sensor, and identifying the initial output constraint type of the system to obtain an unconstrained new variable. S4, constructing a dual neural network combining a shared network and an exclusive network, wherein the shared network learns general dynamics characteristics of the cross-domain aircraft, adopts the same network structure and initial weight, combines local gradient and distributed consistency items, learns cross-domain commonality, and the exclusive