CN-121979262-A - Spacecraft attitude stability control method based on liquid neural network

Abstract

The invention discloses a spacecraft attitude stability control method based on a liquid neural network. The method comprises the following steps of S1, establishing a spacecraft attitude dynamics model under multi-source complex disturbance, S2, designing a conventional spacecraft attitude stabilization controller, combining the attitude dynamics model to construct a data set, S3, training and obtaining a liquid neural network disturbance observer by adopting a liquid time constant network and a neural circuit strategy framework, S4, designing the spacecraft attitude stabilization controller based on the liquid neural network disturbance observer, and proving the stability of the closed loop system by using a Lyapunov function, S5, verifying the estimation effect of the liquid neural network disturbance observer and the stability effect of a spacecraft attitude control system through simulation, and the result shows that the method has the following effects of effectively estimating the disturbance of the spacecraft attitude and realizing the stable control of the spacecraft attitude, has high real-time, low power consumption and strong robustness, and is beneficial to improving the stable control efficiency of the spacecraft attitude.

Inventors

• LIU CHUANG
• MA JINGCHUN
• YUE XIAOKUI
• KANG JUNJIE
• LONG JIAYI
• LI LI

Assignees

• 西北工业大学

Dates

Publication Date

20260505

Application Date

20260202

Claims (1)

1. 1. The spacecraft attitude stability control method based on the liquid neural network is characterized by comprising the following steps of: S1, building a spacecraft attitude dynamics model under multi-source complex disturbance of model parameter uncertainty, external disturbance and actuator fault; S2, designing a conventional spacecraft attitude stabilization controller under the multi-source complex disturbance, combining an attitude dynamics model to obtain error, control and interference information, and integrating the information to construct a data set; S3, adopting a liquid time constant network and a neural circuit strategy architecture, and utilizing the data set to train a liquid neural network interference observation model to construct a liquid neural network interference observer; S4, designing a spacecraft attitude stabilization controller based on a liquid neural network interference observer, and proving the stability of the closed loop system by using a Lyapunov function; S5, verifying the estimation effect of the liquid neural network interference observer and the stabilization effect of the spacecraft attitude control system through simulation; The method comprises the following steps of establishing a spacecraft attitude dynamics model in the step S1, wherein under the multi-source complex disturbance of model parameter uncertainty, external disturbance and actuator fault, the spacecraft attitude angle, attitude angular speed, external disturbance moment, inertia parameter, control moment and unmodeled error upper limit are considered, and the spacecraft attitude dynamics model is established; the data set is obtained under the conventional spacecraft attitude control system in the step S2, and the specific process is that a conventional spacecraft attitude stabilization controller is firstly designed under the multi-source complex disturbance of model parameter uncertainty, external disturbance and actuator fault, the controller is used for realizing spacecraft attitude stabilization control, an attitude dynamics model is combined, a control process is recorded, an attitude angle error, an attitude angle speed error, a control moment and random disturbance information are obtained, and the data set is constructed by integrating the above information; The construction of the liquid neural network disturbance observer in the step S3 comprises the specific processes of adopting a liquid time constant network and a neural circuit strategy framework, taking an attitude angle error, an attitude angle speed error and a control moment in a data set as input, taking a disturbance estimated value as output, training a liquid neural network disturbance observation model, and constructing the liquid neural network disturbance observer based on the model, wherein the disturbance observer is characterized by continuous time dynamics of the liquid time constant network and sparse connection of the neural circuit strategy, the neural circuit strategy is a wiring mode inspired by biological nerves, integrating a plurality of liquid time constant network neurons to be applied to a multi-layer structure, realizing sparse connection among sensory neurons, neuron interlayers, command neurons and motion neuron layers, the liquid time constant network is a dynamic neural network constructed according to a differential equation, and is characterized in that the adaptability of biological synapses is simulated by means of dynamically adjusting time constants, the characteristic is based on STDP rules, the characteristic is a bionic learning rule simulating the strength of synapse in a biological brain, has a dynamic adjustment mechanism, can update the state of synapse with time change, adjusts the connection weights between the neurons in real time, namely, the real-time change of the weighting is adjusted according to the estimated disturbance dynamic disturbance values, and the real-time disturbance values are input, and the mathematical disturbance values are estimated and the disturbance values are real-time: In the formula, As synaptic weights, the variation depends on the time sequence and time difference of the front and back neuron pulses, In order to be a difference in the time between the front and rear pulses, For learning the rate parameter, for controlling the magnitude of the emphasis fade, Determining an influence time range for the time window constant; The liquid neural network model has memory characteristics, the characteristics are based on dynamic neurons of differential equations, through simulating continuous time dynamic behaviors of real neurons, by combining the accumulated effect of historical data along with time and the dynamic adjustment prediction result of current input data, the liquid neural network adjusts neuron state vectors and connection weights in real time by receiving new interference input, a nonlinear activation function is adopted, interference corresponding to spacecraft attitude errors is estimated by using a memory mechanism, a dynamic adjustment mechanism and a recursive feedback mechanism of the model, and the liquid neuron state equation is as follows: In the formula, Is a vector of states of the neurons, Is a time constant which is a function of the time constant, Is an input weight matrix, is adjusted on line by STDP rules, Is an input signal which is provided with a signal, Is a matrix of recursive weights and, Is a bias term that is used to determine, A nonlinear activation function; The spacecraft attitude stabilization controller based on the liquid neural network interference observer is designed in the step S4, and the stability of the closed loop system is proved by Lyapunov function, and the specific process is as follows: inputting an interference estimated value based on a liquid neural network interference observer to the front end of a controller to introduce equivalent compensation, reducing the influence of comprehensive disturbance, and proving the stability of the system through a Lyapunov function; the estimation effect of the liquid neural network interference observer and the stabilization effect of the attitude stabilization system are verified through simulation in the step S5, and the specific process is that random interference is input to the spacecraft attitude stabilization system, the liquid neural network interference observer obtains an interference estimation value according to attitude angle errors, attitude angle speed errors and historical control moment input of the spacecraft, and equivalent compensation is introduced into a controller according to the estimation value so as to verify that the interference estimation and actual interference errors of the observer are smaller, and the system always keeps a stable state.

Description

Spacecraft attitude stability control method based on liquid neural network Technical Field The invention belongs to the technical field of spacecraft control, relates to a spacecraft attitude control method, and in particular relates to a spacecraft attitude stability control method based on a liquid neural network. Background With the continuous expansion of space exploration, the rapid development of the aerospace technology is realized, the on-orbit task of the spacecraft is more complex, and the performance requirement on the attitude control system is continuously improved. However, spacecraft attitude control systems face challenges such as model parameter unknowns, state information unknowns, actuator failures, computational resource limitations, and the like. These factors severely impact the performance of conventional control methods and may even lead to task failure. In view of the above problems, many modern attitude control methods have been proposed, including adaptive control, sliding mode control, H-infinity control, and the like, in which a control method based on an interference observer is one of the very promising approaches in spacecraft attitude control. However, conventional disturbance observers are also unable to estimate external disturbances with complete accuracy, and the corresponding speeds of these methods are relatively slow. Although deep neural networks, such as RNN, transformer, show advantages in some scenes, there are many defects, such as difficulty in meeting real-time requirements of control loops due to high reasoning delay of GPU deployment, incapability of modifying weights once training is finished, difficulty in self-adaptively adjusting internal states, incapability of adapting to dynamic changes on line, requirement of ground retraining if task requirements change, large power consumption of hardware equipment on which the traditional neural network depends, power supply capability of a far-beyond spacecraft, no memory function and limited adaptability to dynamic systems, and similarity of the traditional neural network to a black box, and incapability of being one of important reasons for limiting application of the traditional neural network to the field of aircrafts. The hydrodynamic neural network based on the sparse connection structure has time sequence information processing and dynamic memory capacity by introducing continuous time dynamics and recursion connection structures, and provides a new idea for intelligent control of a complex dynamic system. As a novel neural network structure, the weight variability of the neural network structure makes the neural network structure excellent in processing time-varying and dynamic data, can adapt to continuously-changing environments, and can complete complex tasks through fewer neurons, so that the computational energy consumption is low, and the interpretability of the neural network is greatly improved by virtue of the simple structure of the neural network structure. The liquid neural network adjusts the state vector and the connection weight of the neurons in real time by receiving new interference input, adopts a nonlinear activation function, and analyzes the spacecraft attitude correction strategy by using a memory mechanism, a dynamic adjustment mechanism and a recursive feedback mechanism of a model, so that the liquid neural network has incomparable advantages in the field of spacecraft control by virtue of high efficiency and low energy consumption. Disclosure of Invention The invention aims to make up the defects of the prior art and provides a spacecraft attitude stability control method with high self-adaptability and low power consumption based on a liquid neural network. By the method, stable and self-adaptive control of the attitude of the spacecraft can be realized under the condition of external interference factors, modeling precision requirements of traditional control and high energy consumption defects of a traditional neural network are obviously reduced, requirements of high precision, dynamic adjustment and self-adaptive control of the spacecraft are met, the attitude control efficiency of the spacecraft is improved by assistance, and the service life of an on-orbit satellite is prolonged. The invention is realized by adopting the following technical scheme: S1, building a spacecraft attitude dynamics model under multi-source complex disturbance of model parameter uncertainty, external disturbance and actuator fault; S2, designing a conventional spacecraft attitude stabilization controller under the multi-source complex disturbance, combining an attitude dynamics model to obtain error, control and interference information, and integrating the information to construct a data set; S3, training a liquid neural network interference observation model by using the data set by adopting a liquid time constant network and a neural circuit strategy architecture, and constructing a liquid neu