Sliding mode fault-tolerant neural network control method for flexible satellite based on disturbance observer and fault estimator

Abstract

The present invention discloses a sliding mode fault-tolerant neural network control method for a flexible satellite based on an interference observer and a fault estimator, including: establishing a single-axis rotating flexible satellite dynamics model with actuator faults and internal disturbances; using RBFNNs to approximate system uncertainty; designing an interference observer to estimate the composite interference composed of the neural network approximation error and the internal interference caused by the vibration of the flexible component; designing a fault estimator to obtain actuator fault information; proving the stability of the interference observer and the fault estimator; designing a sliding mode fault-tolerant controller to achieve stabilization control of the satellite attitude; and finally selecting control parameters based on the Lyapunov stability theory to ensure that the system error signal is uniformly bounded. The present invention realizes that a satellite system with flexible component vibration in a complex space environment can maintain steady-state performance in the presence of an actuator fault, and realizes fault-tolerant control of a flexible satellite with actuator faults and system uncertainty.

Description

Technical Field

[0001] The invention relates to a spacecraft control technology, and in particular to a flexible satellite sliding mode fault-tolerant neural network control method based on a disturbance observer and a fault estimator. Background Art

[0002] In order to meet the complex space missions, the types and quantities of flexible accessories carried by satellites are increasing. During the satellite attitude motion, the coupling between the central rigid body and the flexible accessories will excite the vibration of the flexible accessories, causing unexpected satellite attitude motion. Since the damping of the space environment is too small, the vibration of the flexible accessories may even cause the satellite to lose control. Secondly, the satellite operating environment is in space, and the harsh environment may cause execution failures. In addition, there are situations such as uncertainty in the moment of inertia of flexible satellites in orbit. Satellites are of great significance both in military and civilian use. Therefore, studying the robust fault-tolerant control technology of flexible satellites to ensure the stable operation of flexible satellites in orbit is an important research direction for flexible satellites.

[0003] In recent years, the fault-tolerant control problem of flexible satellites has gradually become a hot topic among scholars. Various classical control methods and nonlinear control methods are used to achieve fault-tolerant control of satellite attitude. However, with the continuous development of computer technology, neural networks are gradually used in research in various fields. Neural networks have self-learning, self-organization and self-adaptation capabilities, which enable the network to approximate uncertain systems with arbitrary accuracy. This feature is exactly what is needed to solve the uncertainty problem of flexible satellites. Therefore, the study of flexible satellite fault-tolerant control methods combined with neural networks has important practical significance. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a flexible satellite sliding mode fault-tolerant neural network control method based on a disturbance observer and a fault estimator, so as to improve the orbital steady-state performance of a single-axis rotating flexible satellite in the presence of actuator failures, internal interference and system uncertainty.

[0005] Technical solution: The flexible satellite sliding mode fault-tolerant neural network control method based on interference observer and fault estimator of the present invention comprises the following steps:

[0006] Based on the idea of ​​mechanism modeling, a kinematic model of a single-axis rotating flexible satellite with actuator failure and internal interference is established.

[0007] The RBF neural network is used to approximate the system uncertainty, and the system uncertainty is added into the kinematic model of the single-axis rotating flexible satellite with actuator failure and internal disturbance, thus obtaining the single-axis rotating flexible satellite model with actuator failure, internal disturbance and system uncertainty.

[0008] Design a disturbance observer to estimate the composite disturbance consisting of the internal disturbance caused by the vibration of the flexible component and the estimation error of the neural network weights;

[0009] Design a fault estimator for satellite actuator failures to estimate the impact of the failure on the satellite body;

[0010] The sliding surface is selected and a fault-tolerant controller is designed according to the sliding mode control theory to suppress the influence of actuator failure on the single-axis rotation flexible satellite model.

[0011] Furthermore, after designing the disturbance observer and the fault estimator, the method further includes: selecting a Lyapunov function, selecting parameters to be designed for the disturbance observer and the fault estimator according to the Lyapunov stability theory, and ensuring the stability of the disturbance observer and the fault estimator;

[0012] After designing the error controller, the method further includes: selecting a Lyapunov function of the control system, selecting control parameters according to the Lyapunov stability theory, and ensuring that the system error signal is bounded.

[0013] Furthermore, the kinematic model of the single-axis rotating flexible satellite with actuator failure and internal interference is:

[0014]

[0015] Where, J represents the moment of inertia of the flexible satellite, represents the attitude angular acceleration, G represents the rigid-flexible coupling coefficient, G T represents the transpose of the rigid-flexible coupling coefficient, μ represents the flexible mode, represents the first-order derivative of the flexible mode, represents the second-order derivative of the flexible mode, Λ represents the known stiffness matrix, C m represents the known modal damping matrix, F represents the actuator fault, and u represents the input control torque;

[0016] The kinematic model of the single-axis rotating flexible satellite with actuator failure and internal interference is transformed into the following state space form:

[0017]

[0018] in, represents the internal disturbance caused by the vibration of the flexible component; χ represents the system state vector, including the attitude angle α and the attitude angular velocity

[0019] Furthermore, the RBF neural network is used to approximate the system uncertainty △(χ) as:

[0020]

[0021]

[0022] Among them, the system state vector χ is used as the input vector of the RBF neural network; represents the transpose of the neural network weight estimate; φ(χ)=[φ1(χ) φ2(χ) … φ n (x)] T is the Gaussian basis function vector; ε is the RBF neural network weight estimation error; is the estimated value of △(χ);

[0023] The system uncertainty △(χ) is added to the state space form of the kinematic model of the single-axis rotating flexible satellite with actuator failure and internal disturbance, and the single-axis rotating flexible satellite model with actuator failure, internal disturbance and system uncertainty is obtained as follows:

[0024]

[0025] Furthermore, the design method of the disturbance observer is:

[0026] According to the single-axis rotating flexible satellite model with actuator failure, internal disturbance and system uncertainty, the disturbance observer is designed as:

[0027]

[0028] Where D is the composite disturbance, which represents the internal disturbance D caused by the vibration of the flexible component. g and the minimum approximation error of the neural network ε * The composite value of is the estimated value of the composite interference D; Q D is the intermediate variable of the disturbance observer; Indicates Q D The first-order derivative of ; N1 is the disturbance observer parameter to be designed; J represents the moment of inertia of the flexible satellite, G represents the rigid-flexible coupling coefficient, and α and represent the attitude angle and attitude angular velocity of the flexible satellite respectively, is the estimated value of the actuator fault F; Represents the transpose of the neural network weight estimates.

[0029] Furthermore, the fault observer is designed as follows:

[0030]

[0031] in, is the estimated value of the actuator fault F; Q F is the intermediate variable of the fault observer; Indicates QF The first-order derivative of ; N2 is the fault observer parameter to be designed; χ is the system state vector; φ(χ) is the Gaussian basis function vector of the neural network; is the transpose of the neural network weight vector estimate; J represents the moment of inertia of the flexible satellite, G represents the rigid-flexible coupling coefficient, is the estimated value of the composite interference D;

[0032] Furthermore, the fault-tolerant controller is designed as follows:

[0033]

[0034] Among them, u * represents the controller output of the preliminary design; a is a positive definite diagonal matrix with appropriate dimension to be designed in the sliding surface s = aχ, and it must satisfy the existence of Moore-Penrose pseudo-inverse of the aB matrix; ρ>0 is the reaching law The parameters to be designed in , sgn(s)=[sgn(s1) sgn(s2)] T is the symbol function vector, sgn(s i ) is the symbol function, s i ,i=1,2 represents the i-th item of the sliding surface s; is a symmetric Sigmoid function vector, β is the parameter to be designed, β>0, J represents the moment of inertia of the flexible satellite, G represents the rigid-flexible coupling coefficient, is the estimate of the composite interference D, is the estimated value of the actuator fault F, is the transpose of the neural network weight vector estimate; k is the controller parameter to be designed, k>0.

[0035] The invention discloses a flexible satellite sliding mode fault-tolerant neural network control system based on an interference observer and a fault estimator. The controlled object of the control system is a flexible satellite model, and the control quantity is an actuator fault. The control system comprises an RBF neural network, a fault estimator, an interference observer and a fault-tolerant controller. The RBF neural network approximates system uncertainty according to state variables output by the flexible satellite model, and uses the RBF neural network weight estimation error to compensate the flexible satellite model. The estimated value of the RBF neural network weight vector is used as the input of the fault estimator, the interference observer and the fault-tolerant controller. The fault estimator and the interference observer are integrated with each other, and the actuator fault estimation value output by the fault estimator and the composite interference estimation value output by the interference observer are both used as the input of the fault-tolerant controller. The input control torque output by the fault-tolerant controller is used to suppress the influence of the actuator fault of the flexible satellite on the flexible satellite.

[0036] The flexible satellite sliding mode fault-tolerant neural network control system based on interference observer and fault estimator of the present invention comprises:

[0037] The model building module is used to build a single-axis rotating flexible satellite model with actuator failure and internal interference based on the mechanism modeling concept;

[0038] RBF neural network is used to approximate system uncertainty and add it into the kinematic model of a single-axis rotating flexible satellite with actuator failure and internal disturbance;

[0039] A disturbance observer is used to estimate the impact of the combined disturbance consisting of the internal disturbance caused by the vibration of the flexible component and the error in the estimation of the neural network weights on the system;

[0040] A fault estimator, used for estimating the fault information of the flexible satellite actuator;

[0041] The disturbance observer and fault estimator stability analysis module selects the Lyapunov function and selects the disturbance observer and fault estimator parameters according to the Lyapunov stability theory to ensure that the disturbance observer and fault estimator are stable and can effectively estimate the disturbance and fault information;

[0042] Fault-tolerant controller, based on the designed flexible satellite fault-tolerant controller, the attitude angle and attitude angular velocity of the single-axis rotating satellite are controlled to achieve fault-tolerant control, and remain stable after a fault occurs;

[0043] The effectiveness analysis module selects the Lyapunov function of the control system and chooses the control parameters according to the Lyapunov stability theory to ensure the boundedness of the system error signal.

[0044] A device of the present invention includes a memory and a processor, wherein:

[0045] A memory for storing computer programs that can be run on the processor;

[0046] The processor is used to execute the steps of the flexible satellite sliding mode fault-tolerant neural network control method based on the interference observer and the fault estimator when running the computer program.

[0047] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: an interference observer and a fault estimator are designed to specifically and accurately grasp the interference and fault information, which is convenient for compensating for interference and faults in subsequent controller design to weaken their influence; a neural network is used to approximate system uncertainty, and the self-learning, self-organizing and adaptive capabilities of the neural network, as well as the characteristics of approximating nonlinearity with arbitrary precision, are utilized to obtain estimated information of system uncertainty and reduce system uncertainty; the controller design uses a sliding mode control method, which is widely used in engineering due to its strong robustness. In the sliding mode control, a symmetric Sigmoid function vector is used to replace the sign function to avoid the jitter phenomenon that may occur in practical applications, thereby realizing robust fault-tolerant control of flexible satellites. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a system structure block diagram of the present invention;

[0049] Figure 2 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0050] The present invention will be further described below in conjunction with the accompanying drawings.

[0051] like Figure 1 As shown, the controlled object of the control system is the flexible satellite model, and the controlled variable is the actuator fault. The control system includes an RBF neural network, a fault estimator, a disturbance observer and a fault-tolerant controller. The RBF neural network approximates the system uncertainty according to the state variables output by the flexible satellite model, and uses the RBF neural network weight estimation error to compensate the flexible satellite model, and the estimated value of the RBF neural network weight vector is used as the input of the fault estimator, the disturbance observer and the fault-tolerant controller. The fault estimator and the disturbance observer are integrated with each other, and the actuator fault estimation value output by the fault estimator and the composite disturbance estimation value output by the disturbance observer are both used as the input of the fault-tolerant controller. The input control torque output by the fault-tolerant controller is used to suppress the influence of the actuator fault of the flexible satellite on the flexible satellite.

[0052] The flexible satellite sliding mode fault-tolerant neural network control method based on the interference observer and fault estimator of the present invention includes: establishing a kinematic model of a single-axis rotating flexible satellite with actuator faults and internal interference; using RBFNNs to approximate system uncertainty; regarding the neural network approximation error and the internal interference caused by the vibration of the flexible component as a composite interference, and designing an interference observer for estimation; redesigning the fault estimator to obtain actuator fault information; proving the stability of the interference observer and the fault estimator; designing a sliding mode fault-tolerant controller based on the estimated information obtained by the interference observer and the fault estimator to achieve stabilization control of the satellite attitude; and finally selecting control parameters based on the Lyapunov stability theory to ensure that the system error signal is uniformly bounded. The present invention enables a satellite system with flexible component oscillations in a complex space environment to maintain steady-state performance in the presence of actuator faults, and achieves fault-tolerant control of a flexible satellite with actuator faults and system uncertainty. Figure 2 As shown, the specific steps include:

[0053] (1) Based on the idea of ​​mechanism modeling, a kinematic model of a single-axis rotating flexible satellite with actuator failure and internal disturbance is established;

[0054] Specifically, the kinematic model of a single-axis rotating flexible satellite with actuator failure and internal interference is as follows:

[0055]

[0056] Where, J represents the moment of inertia of the flexible satellite; α represents the attitude angle of the flexible satellite; G represents the rigid-flexible coupling coefficient; μ represents the flexible mode, represents the first-order derivative of the flexible mode μ, represents the second-order derivative of the flexible mode μ; Λ represents the known stiffness matrix, satisfying C m represents the known modal damping matrix, satisfying C m =[diag(2ξ i ω i ),i=1,2,...,n],ω i is the natural frequency of the mode, ξ i represents the damping ratio of the mode, n represents the modal order, F represents the actuator fault, and u represents the controller output torque.

[0057] The dynamic model is transformed into a state space form, and the system uncertainty is added, which is as follows:

[0058]

[0059] in, represents the internal disturbance caused by the vibration of the flexible component; χ represents the system state vector, which includes the attitude angle α and attitude angular velocity of the flexible satellite F represents actuator fault; u represents input control torque, and △(χ) is the system uncertainty term.

[0060] Note 1: According to the physical characteristics of the flexible satellite, when the satellite moves in space, any element in the rigid-flexible coupling coefficient matrix G is much smaller than the value of the element in the moment of inertia J of the flexible satellite, that is, IG T J -1 G and J-GG T It's very strange.

[0061] Note 2: It is regarded as the internal disturbance caused by the vibration of the flexible component, which is bounded and can be estimated by designing a disturbance observer.

[0062] (2) Neural networks, used to approximate system uncertainty;

[0063] Specifically, since RBF neural networks (RBFNNs) have the ability to approximate continuous nonlinear mappings, the system uncertainty △(χ) and its estimated value are expressed by RBFNNs as follows:

[0064]

[0065]

[0066] Among them, the system state vector χ∈R 2×1 As the input vector of RBFNNs; is the transpose of the neural network weight estimate; φ(χ)=[φ1(χ) φ2(χ) … φ n (x)] T is a Gaussian basis function vector satisfying ||φ(χ)||≤c φ , that is, c φ is the upper bound of φ(χ); ε is the RBF neural network weight estimation error, satisfying Right now is the upper bound of the approximation error; is an estimate of △(χ).

[0067] In the tight set Ω χ ∈R 2×1 In this paper, RBFNNs can approximate the system uncertainty △(χ) with arbitrarily small accuracy, that is,

[0068] △(χ)=w *T φ(χ)+ε * (5)

[0069] Among them, w* is the optimal weight matrix, and w * The upper bound of * is the minimum approximation error.

[0070] The Gaussian basis radial function has the following form:

[0071]

[0072] Among them, c i ,σ i are the center and width of the Gaussian radial basis function of the i-th layer respectively.

[0073] The optimal weight matrix of RBFNNs is defined as:

[0074]

[0075] Among them, M χ is the set of state vectors; Ω △ is the feasible domain of weight parameters; Represents an estimate of the weights of a neural network; Represents the uncertainty obtained under the estimated weights of the neural network; under the definition of the optimal weight matrix, we can get:

[0076]

[0077] (3) A disturbance observer is used to estimate the impact of the combined disturbance caused by the internal disturbance caused by the oscillation of the flexible component and the error in the estimation of the neural network weights on the system;

[0078] Specifically, for the nonlinear system described by formula (2), we first define D as the internal disturbance D caused by the vibration of the flexible component: g and the minimum approximation error ε * The composite value of

[0079] D=BD g +ε * (9)

[0080] in, J represents the moment of inertia of the flexible satellite; G represents the rigid-flexible coupling coefficient.

[0081] Assumption 1: The actuator fault F is bounded and its derivative satisfies the condition is a normal number.

[0082] Assumption 2: is bounded, and ε * is bounded, D is bounded, and is a normal number.

[0083] Assumption 3: ||φ(χ)||≤c φ , that is, c φ is an upper bound of φ(χ).

[0084] Assumption 4: ε is the approximation error vector of RBFNNs, satisfying Right now is the upper bound of the approximation error.

[0085] The disturbance observer is designed as follows:

[0086]

[0087] in, is the estimated value of the composite interference D; Q D is the intermediate variable of the disturbance observer; N1 is the disturbance observer parameter to be designed.

[0088] (4) a fault estimator, used to estimate the fault information of the flexible satellite actuator;

[0089] Specifically, the fault estimator is designed as follows:

[0090]

[0091] in, is the estimated value of the actuator fault F; Q F is the intermediate variable of the fault estimator; N2 is the fault estimator parameter to be designed.

[0092] (5) The disturbance observer and fault estimator stability analysis module selects the Lyapunov function and selects the disturbance observer and fault estimator parameters according to the Lyapunov stability theory to ensure that the disturbance observer and fault estimator are stable and can effectively estimate the disturbance and fault information;

[0093] Specific, definition is the estimation error of the disturbance observer, is the estimation error of the fault estimator, is the neural network weight estimation error. and fault estimation error The derivatives are:

[0094]

[0095] in, represents the first derivative of the estimation error of the composite disturbance; represents the first derivative of the composite disturbance; represents the first derivative of the estimate of the composite disturbance D; represents the first-order derivative of the intermediate variable of the disturbance observer; Represents the first-order derivative of the state quantity.

[0096]

[0097] in, The first derivative of the estimation error representing the actuator fault; The first derivative representing the actuator fault; represents the estimated first-order derivative of the actuator fault F; represents the first derivative of the intermediate variable of the fault estimator; Represents the first-order derivative of the state quantity.

[0098] The stability of the disturbance observer and fault estimator is analyzed below:

[0099] Take the Lyapunov function as follows:

[0100]

[0101] Where V1 represents the Lyapunov function of the disturbance observer and fault estimator;

[0102] Derivation of formula (14) yields:

[0103]

[0104] in, is the upper bound of the first-order derivative of the fault F(t), is the upper bound of the first-order derivative of the composite disturbance D(t), c φ is the upper bound of the neural network Gaussian radial basis function φ(χ), ||φ(χ)||≤c φ , c φ >0; α1, α2, α3, α4 are the scaling parameters to be designed; I is the identity matrix with appropriate dimensions; choose appropriate parameters α1, α2, α3, α4 and N1, N2 so that α1||N1|| 2 +2N1+(α3-α2-1)I and Positive definiteness ensures that the disturbance observer and fault estimator are stable.

[0105] (6) Fault-tolerant controller: According to the designed flexible satellite fault-tolerant controller, the attitude angle and attitude angular velocity of the single-axis rotating satellite are controlled to achieve fault-tolerant control, and the controller remains stable after a fault occurs.

[0106] Specifically, select the sliding surface:

[0107] s=aχ (16)

[0108] Where a is a positive definite diagonal matrix with appropriate dimension to be designed, and it must satisfy the existence of Moore-Penrose pseudo-inverse of aB matrix. Deriving equation (16) yields:

[0109]

[0110] Among them, w T Represents the transpose of the neural network weight matrix.

[0111] The sliding surface is approached by the constant velocity approach law, and the specific form is as follows:

[0112]

[0113] Where ρ>0 is the parameter to be designed; sgn(s)=[sgn(s1) sgn(s2)] T is the symbol function vector, sgn(s i ) is the symbol function, s i ,i=1,2 represents the i-th item of vector s.

[0114] Design the satellite closed-loop system controller with faults and internal interference as follows:

[0115]

[0116] Among them, k is the parameter to be designed, k>0.

[0117] In order to eliminate the chattering phenomenon caused by the sign function in the controller, the symmetric Sigmoid function vector is used to replace the sign function vector in the controller. The symmetric Sigmoid function vector is taken as:

[0118]

[0119] Among them, β is the parameter to be designed, β>0. The modified controller is:

[0120]

[0121] because and the sign function sgn(s) are bounded, so the difference between the two is is also bounded, so

[0122]

[0123] (7) Effectiveness analysis module: selects the Lyapunov function of the control system and chooses the control parameters according to the Lyapunov stability theory to ensure the boundedness of the system error signal.

[0124] Specifically, the constructed system Lyapunov equation is:

[0125]

[0126] Where V2 represents the Lyapunov function of the control system; Γ is the positive definite symmetric matrix to be designed; Represents the neural network weight estimation error; Represents the transpose of the neural network weight estimation error.

[0127] By taking the derivative of both sides of equation (23), we can obtain:

[0128]

[0129] Will Substituting into formula (24) we get:

[0130]

[0131] The adaptive law of neural network weights is taken as:

[0132]

[0133] Among them, γ is the parameter to be designed, γ>0; Γ -1 represents the inverse matrix of Γ.

[0134] Substituting formula (26) into formula (25), we obtain:

[0135]

[0136] According to the definition of neural network weight estimation error, we have:

[0137]

[0138] Substituting formula (28) into formula (27), we get:

[0139]

[0140] in, So the system is eventually uniformly bounded.

[0141] The flexible satellite sliding mode fault-tolerant neural network control system based on a disturbance observer and a fault estimator of the present invention comprises:

[0142] Model building module, which is used to establish the kinematic model of a single-axis rotating flexible satellite with actuator failure and internal interference based on the mechanism modeling concept;

[0143] RBF neural network is used to approximate system uncertainty and add it into the kinematic model of a single-axis rotating flexible satellite with actuator failure and internal disturbance;

[0144] A disturbance observer is used to estimate the impact of the combined disturbance consisting of the internal disturbance caused by the oscillation of the flexible component and the error in the estimation of the neural network weights on the system;

[0145] A fault estimator, used for estimating the fault information of the flexible satellite actuator;

[0146] The disturbance observer and fault estimator stability analysis module selects the Lyapunov function and selects the disturbance observer and fault estimator parameters according to the Lyapunov stability theory to ensure that the disturbance observer and fault estimator are stable and can effectively estimate the disturbance and fault information;

[0147] Fault-tolerant controller, based on the designed flexible satellite fault-tolerant controller, the attitude angle and attitude angular velocity of the single-axis rotating satellite are controlled to achieve fault-tolerant control, and remain stable after a fault occurs;

[0148] The effectiveness analysis module selects the Lyapunov function of the control system and chooses the control parameters according to the Lyapunov stability theory to ensure the boundedness of the system error signal.

[0149] A device of the present invention includes a memory and a processor, wherein:

[0150] A memory for storing computer programs that can be run on the processor;

[0151] The processor is used to execute the steps of the flexible satellite sliding mode fault-tolerant neural network control method based on the interference observer and the fault estimator when running the computer program, and achieve the technical effect consistent with the above method.

[0152] A storage medium of the present invention stores a computer program, which, when executed by at least one processor, implements the steps of the flexible satellite sliding mode fault-tolerant neural network control method based on an interference observer and a fault estimator, and achieves the same technical effect as the above method.

[0153] In summary, the present invention can ensure the normal operation of the single-axis rotating flexible satellite in orbit in the presence of actuator failure, system uncertainty and internal disturbance, and realizes the attitude robust fault-tolerant control of the single-axis rotating flexible satellite under actuator failure.

Claims

1. A flexible satellite sliding mode fault-tolerant neural network control method based on disturbance observer and fault estimator, characterized in that: The following steps are involved: Based on the idea of ​​mechanism modeling, the kinematic model of a single-axis rotating flexible satellite with actuator failure and internal interference is established: Where, J represents the moment of inertia of the flexible satellite, represents the attitude angular acceleration, G represents the rigid-flexible coupling coefficient, G T represents the transpose of the rigid-flexible coupling coefficient, μ represents the flexible mode, represents the first-order derivative of the flexible mode, represents the second-order derivative of the flexible mode, Λ represents the known stiffness matrix, C m represents the known modal damping matrix, F represents the actuator fault, and u represents the input control torque; The kinematic model of the single-axis rotating flexible satellite with actuator failure and internal interference is transformed into the following state space form: in, represents the internal disturbance caused by the vibration of the flexible component; χ represents the system state vector, including the attitude angle α and the attitude angular velocity The RBF neural network is used to approximate the system uncertainty, and the system uncertainty is added into the kinematic model of the single-axis rotating flexible satellite with actuator failure and internal disturbance, thus obtaining the single-axis rotating flexible satellite model with actuator failure, internal disturbance and system uncertainty. The RBF neural network is used to approximate the system uncertainty △(χ) as follows: Among them, the system state vector χ is used as the input vector of the RBF neural network; represents the transpose of the neural network weight estimate; φ(χ)=[φ1(χ)φ2(χ)…φ n (x)] T is the Gaussian basis function vector; ε is the RBF neural network weight estimation error; is the estimated value of △(χ); The system uncertainty △(χ) is added to the state space form of the kinematic model of the single-axis rotating flexible satellite with actuator failure and internal disturbance, and the single-axis rotating flexible satellite model with actuator failure, internal disturbance and system uncertainty is obtained as follows: A disturbance observer is designed to estimate the composite disturbance consisting of the internal disturbance caused by the vibration of the flexible component and the neural network weight estimation error; the design method of the disturbance observer is: According to the single-axis rotating flexible satellite model with actuator failure, internal disturbance and system uncertainty, the disturbance observer is designed as: Where D is the composite disturbance, which represents the internal disturbance D caused by the vibration of the flexible component. g and the minimum approximation error of the neural network ε * The composite value of is the estimated value of the composite interference D; Q D is the intermediate variable of the disturbance observer; Indicates Q D The first-order derivative of ; N1 is the disturbance observer parameter to be designed; J represents the moment of inertia of the flexible satellite, G represents the rigid-flexible coupling coefficient, and α and represent the attitude angle and attitude angular velocity of the flexible satellite respectively, is the estimated value of the actuator fault F; represents the transpose of the neural network weight estimate; A fault estimator is designed for the satellite actuator fault to estimate the impact of the fault on the satellite body; the fault observer is designed in the following form: in, is the estimated value of the actuator fault F; Q F is the intermediate variable of the fault observer; Indicates Q F The first-order derivative of ; N2 is the fault observer parameter to be designed; χ is the system state vector; φ(χ) is the Gaussian basis function vector of the neural network; is the transpose of the neural network weight vector estimate; J represents the moment of inertia of the flexible satellite, G represents the rigid-flexible coupling coefficient, is the estimated value of the composite interference D; The sliding surface is selected and a fault-tolerant controller is designed according to the sliding mode control theory to suppress the influence of actuator failure on the single-axis rotation flexible satellite model. The fault-tolerant controller is designed as follows: Among them, u * represents the controller output of the preliminary design; a is a positive definite diagonal matrix with appropriate dimension to be designed in the sliding surface s = aχ, and it must satisfy the existence of Moore-Penrose pseudo-inverse of the aB matrix; ρ>0 Approach Law The parameters to be designed in , sgn(s)=[sgn(s1)sgn(s2)] T is the symbol function vector, sgn(s i ) is the symbol function, s i ,i=1,2 represents the i-th item of the sliding surface s; is the symmetric Sigmoid function vector, β is the parameter to be designed, β >0 , J represents the moment of inertia of the flexible satellite, G represents the rigid-flexible coupling coefficient, is the estimate of the composite interference D, is the estimated value of the actuator fault F, is the transpose of the neural network weight vector estimate; k is the controller parameter to be designed, k>

0.

2. The flexible satellite sliding mode fault-tolerant neural network control method based on disturbance observer and fault estimator according to claim 1 is characterized in that: After designing the disturbance observer and the fault estimator, the method further includes: selecting a Lyapunov function, selecting parameters to be designed for the disturbance observer and the fault estimator according to the Lyapunov stability theory, and ensuring the stability of the disturbance observer and the fault estimator; After designing the fault-tolerant controller, the method further comprises: selecting a Lyapunov function of the control system, selecting control parameters according to the Lyapunov stability theory, and ensuring the boundedness of the system error signal.

3. A flexible satellite sliding mode fault-tolerant neural network control system based on disturbance observer and fault estimator, characterized in that: The system is used for the control method described in any one of claims 1-2, wherein the controlled object of the control system is a flexible satellite model, the control quantity is an actuator fault, the control system includes an RBF neural network, a fault estimator, an interference observer and a fault-tolerant controller, the RBF neural network approximates system uncertainty according to the state variables output by the flexible satellite model, uses the RBF neural network weight estimation error to compensate the flexible satellite model, and uses the estimated value of the RBF neural network weight vector as the input of the fault estimator, the interference observer and the fault-tolerant controller, the fault estimator and the interference observer are integrated with each other, and the actuator fault estimation value output by the fault estimator and the composite interference estimation value output by the interference observer are both used as the input of the fault-tolerant controller, and the input control torque output by the fault-tolerant controller is used to suppress the influence of the actuator fault of the flexible satellite on the flexible satellite.

4. A flexible satellite sliding mode fault-tolerant neural network control system based on disturbance observer and fault estimator, characterized in that: The system is used for the control method according to any one of claims 1 to 2, comprising: The model building module is used to build a single-axis rotating flexible satellite model with actuator failure and internal interference based on the mechanism modeling concept; RBF neural network is used to approximate system uncertainty and add it into the kinematic model of a single-axis rotating flexible satellite with actuator failure and internal disturbance; A disturbance observer is used to estimate the impact of the combined disturbance consisting of the internal disturbance caused by the vibration of the flexible component and the error in the estimation of the neural network weights on the system; A fault estimator, used for estimating the fault information of the flexible satellite actuator; The disturbance observer and fault estimator stability analysis module selects the Lyapunov function and selects the disturbance observer and fault estimator parameters according to the Lyapunov stability theory to ensure that the disturbance observer and fault estimator are stable and can effectively estimate the disturbance and fault information; Fault-tolerant controller, based on the designed flexible satellite fault-tolerant controller, the attitude angle and attitude angular velocity of the single-axis rotating satellite are controlled to achieve fault-tolerant control, and remain stable after a fault occurs; The effectiveness analysis module selects the Lyapunov function of the control system and chooses the control parameters according to the Lyapunov stability theory to ensure the boundedness of the system error signal.

5. A device, characterized in that: comprising a memory and a processor, wherein: A memory for storing computer programs that can be run on the processor; A processor is used to execute the steps of the flexible satellite sliding mode fault-tolerant neural network control method based on interference observer and fault estimator as described in any one of claims 1-2 when running the computer program.

Images