State estimation method and system for complex field neural network with distributed time delay

Abstract

The invention discloses a state estimation method for a complex field neural network with distributed time delay. The state estimation method comprises the following steps: constructing a distributed time delay complex field neural network model, an elastic state estimator and a state estimation error calculation model under a Round-Robin protocol; analyzing the stability of the state estimation error model according to a Lyapunov stability theory, and when the state estimation error model is in a stable state, calculating a gain matrix related to each sensor of the elastic state estimator when a l2-linfinity estimation performance index is met; and the elastic state estimator estimates a state vector of a neuron in the complex field neural network at the next moment based on a gain matrix related to the current sensor with the transmission authority. An elastic state estimator is designed based on a complex field neural network with distributed time delay of a Round-Robin protocol, when a state estimation error is stable, a gain matrix related to each sensor in the elastic state estimator when a l2-linfinity estimation performance index is met is calculated, and the elastic state estimator can effectively estimate a target state.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of state estimation, and more particularly, the present application relates to a state estimation method and system for a complex domain neural network with distributed time delay. BACKGROUND

[0002] Complex domain neural networks have very wide applications in many fields, such as sensing and imaging of electromagnetic waves and light waves, blur recovery in image processing, etc. In addition, compared with real domain neural networks, complex domain neural networks have obvious advantages in the processing of some problems. For example, a single complex domain neuron can solve the detection of XOR problems and symmetry problems, while a single real domain neuron cannot solve such problems. In most practical applications of neural networks, the full state information of neurons needs to be known in order to reasonably use these state information to achieve the predetermined goal. However, for a neural network composed of a large number of interconnected neurons, the measurement output of the neural network can only obtain partial state information of the neurons, and rarely can completely know the state information of the neurons. In order to know the state information of each neuron in advance, so as to reasonably use these state information in the application of neural networks to achieve the predetermined goal, the state estimation problem of neural networks needs to be solved. At present, in the related research on the state estimation problem of neural networks, most of them are for real domain neural networks, and only a few are for complex domain neural networks which have more complex modeling forms with complex parameters and variables to process complex information. In fact, complex domain neural networks have the potential advantage of having more extensive and important practical applications, and it is of great theoretical significance and practical value to propose new theoretical methods to study the state estimation problem of complex domain neural networks.

[0003] On the other hand, in the research on the state estimation problem of complex domain neural networks, many sensors are considered to measure the complex domain neural network, and the multiple sensors communicate with the remote state estimator through a shared high-speed communication network. Due to the limited bandwidth of the communication network, when all the measurement data of the sensors are transmitted through the communication channel at the same time, data collision will inevitably occur. SUMMARY

[0004] The present application provides a state estimation method for a complex domain neural network with distributed time delay, aiming at improving the above problems.

[0005] The present application is implemented as follows: a state estimation method for a complex domain neural network with distributed time delay, the method is as follows:

[0006] (1) constructing a distributed time-delay complex-valued neural network model under a Round-Robin protocol, and an elastic state estimator of the distributed time-delay complex-valued neural network model under the Round-Robin protocol;

[0007] (2) constructing a state estimation error calculation model based on the distributed time-delay complex-valued neural network model and the elastic state estimator under the Round-Robin protocol;

[0008] (3) analyzing the stability of the state estimation error model according to Lyapunov stability theory, and calculating the gain matrix related to each sensor of the elastic state estimator when the state estimation error model is detected to be in a stable state, so as to meet the l2-l ∞ estimation performance index;

[0009] (4) the elastic state estimator receiving the current detected state information of the sensor with transmission permission, and estimating the state of the neuron in the complex-valued neural network at the next time based on the gain matrix related to the sensor with transmission permission.

[0010] Further, the construction process of the distributed time-delay complex-valued neural network model under the Round-Robin protocol is as follows:

[0011] (11) constructing a complex-valued neural network model with distributed time delay;

[0012] (12) establishing a transmission model of sensor measurement data under the Round-Robin protocol scheduling, the sensor being used to detect the state information of the distributed time-delay complex-valued neural network and transmit to the elastic state estimator;

[0013] (13) establishing a distributed time-delay complex-valued neural network model under the Round-Robin protocol.

[0014] Further, the distributed time-delay complex-valued neural network model under the Round-Robin protocol is as follows:

[0015]

[0016] wherein, is the state vector ζ(k) of the neuron at time k and the actually transmitted measurement information is an augmented vector, is the state vector ζ(k+1) of the neuron at time k+1 and the actually transmitted measurement information is an augmented vector, is the real part μ R (k) and the imaginary part μ I (k) of the interference input μ(k) at time k, and For the augmented parameter matrix, it has the following form:

[0017]

[0018] Further, the elastic state estimator is specified as follows:

[0019]

[0020] wherein, is the state The state estimation at time k, G ρ(k) is the gain matrix related to the sensor with transmission right ρ(k), ΔG ρ(k) (k) represents the uncertainty of the gain matrix G ρ(k) , and ΔG ρ(k) (k) = A ρ(k) Γ(k)B ρ(k) , wherein A ρ(k) and B ρ(k) are known real constant matrices, and Γ(k) is a time-varying matrix function satisfying the condition Γ T (k)Γ(k)≤I.

[0021] Further, the state estimation error calculation model is specified as follows:

[0022]

[0023] Further, the transmission model of the sensor measurement data under the Round-Robin protocol scheduling is specified as follows:

[0024] The actual transmission measurement of the sensor with transmission right at time k is specified as follows:

[0025]

[0026] wherein λ j (k) represents the state of the neuron detected by the jth sensor at time k, and let:

[0027]

[0028] wherein δ(ρ(k)-j) (j∈{1,2,...,m}) represents the Kronecker function, δ(ρ(k)-j)=1 when ρ(k)=j, otherwise δ(ρ(k)-j)=0, and the actual transmission measurement is specified as follows:

[0029]

[0030] wherein, represents the measurement actually transmitted at time k-1, and I is the identity matrix.

[0031] The present invention is implemented as follows: a flexible state estimation system of a complex domain neural network with distributed time delay under a Round-Robin protocol, the system comprising:

[0032] A distributed time-delay complex domain neural network under the Round-Robin protocol, and m sensors for detecting states of neurons in the distributed time-delay complex domain neural network;

[0033] An elastic state estimator, communicating with m sensors;

[0034] A state estimation error calculation model, connected with the elastic state estimator and the distributed time-delay complex domain neural network;

[0035] a processor connected to the state estimation error calculation model;

[0036] m sensors detect the current neuron state of the distributed time-delay complex domain neural network. A sensor that obtains the transmission permission outputs the detected current neuron state to the elastic state estimator. The elastic state estimator predicts the state vector of the neuron in the distributed time-delay complex domain neural network at the next moment and outputs it to the state estimation error calculation model. The state estimation error calculation model calculates the state estimation error at the next moment based on the actual state vector of the distributed time-delay complex domain neural network at the next moment. When the state estimation error is stable, the processor calculates the state estimation error that satisfies l2-l ∞ When estimating the performance index, the gain matrix associated with each sensor in the elastic state estimator is used. Then, the elastic state estimator estimates the state of the neurons in the complex domain neural network at the next moment based on the gain matrix associated with the sensor that currently obtains the transmission permission.

[0037] The present invention designs a complex domain neural network with distributed time delay based on Round-Robin protocol, an elastic state estimator and a corresponding state estimation error calculation model. When the state estimation error is stable, the calculation satisfies l2-l ∞ The performance index is estimated based on the gain matrix associated with each sensor in the elastic state estimator, which can effectively estimate the target state. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of a method for elastic state estimation using a complex domain neural network with distributed time delay provided by an embodiment of the present invention;

[0039] Figure 2State estimation and state estimation error under the real state x1(k) provided by an embodiment of the present invention, where (a) is the state estimation trajectory and (b) is the state estimation error trajectory;

[0040] Figure 3 State estimation and state estimation error under the real state x2(k) provided by an embodiment of the present invention, where (a) is the state estimation trajectory and (b) is the state estimation error trajectory;

[0041] Figure 4 State estimation and state estimation error under the imaginary state y1(k) provided by an embodiment of the present invention, where (a) is the state estimation trajectory and (b) is the state estimation error trajectory;

[0042] Figure 5 State estimation and state estimation error under the imaginary state y2(k) provided by an embodiment of the present invention, where (a) is the state estimation trajectory and (b) is the state estimation error trajectory;

[0043] Figure 6 The real part output provided by the embodiment of the present invention State estimation and state estimation error under , where (a) is the state estimation trajectory and (b) is the state estimation error trajectory;

[0044] Figure 7 The embodiment of the present invention provides an imaginary output State estimation and state estimation error under , where (a) is the state estimation trajectory and (b) is the state estimation error trajectory. DETAILED DESCRIPTION

[0045] The specific implementation methods of the present invention will be further explained in detail below by describing the embodiments with reference to the accompanying drawings, so as to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0046] In order to improve the data conflict problem that occurs when multiple sensors simultaneously transmit data to the state estimator, the present invention uses the Round-Robin communication protocol to reasonably allocate sensors with network usage rights in the communication network, thereby achieving a reasonable allocation of limited communication network resources. When considering the scheduling of the communication protocol, the existing state estimation scheme designed for complex domain neural networks under periodic and complete measurement information is no longer applicable. The present invention provides a flexible state estimation system for complex domain neural networks with distributed time delays under the Round-Robin protocol, which includes:

[0047] A complex domain neural network, and m sensors for detecting states of neurons in the complex domain neural network;

[0048] an elastic state estimator, in communication connection with the m sensors;

[0049] a state estimation error calculation model, connected with the elastic state estimator and the complex domain neural network model;

[0050] a processor, connected with the state estimation error calculation model;

[0051] The m sensors detect the current neuron state of the complex domain neural network, and the sensor with the transmission permission outputs the detected current neuron state to the elastic state estimator, the elastic state estimator predicts the state vector of the neuron in the complex domain neural network at the next moment, and outputs it to the state estimation error calculation model, the state estimation error calculation model calculates the state estimation error at the next moment based on the actual state vector of the complex domain neural network model at the next moment, and when the state estimation error is stable, the processor calculates the gain matrix related to each sensor in the elastic state estimator when the l2-l ∞ estimation performance index is met, and the elastic state estimator estimates the state of the neuron in the complex domain neural network at the next moment based on the gain matrix related to the sensor with the current transmission permission.

[0052] Figure 1 A flowchart of an elastic state estimation method for a complex domain neural network with distributed time delay provided by the embodiment of the present application, the method comprises:

[0053] (1) constructing a distributed time delay complex domain neural network model under the Round-Robin protocol, and an elastic state estimator of the distributed time delay complex domain neural network model;

[0054] (2) constructing a state estimation error calculation model based on the distributed time delay complex domain neural network model and the elastic state estimator;

[0055] (3) analyzing the stability of the state estimation error model according to the Lyapunov stability theory, and when it is detected that the state estimation error model is in a stable state, calculating the gain matrix related to each sensor in the elastic state estimator when the l2-l ∞ estimation performance index is met;

[0056] (4) the elastic state estimator receives the state information detected by the sensor with the transmission permission, and estimates the state of the neuron in the complex domain neural network at the next moment based on the gain matrix related to the sensor with the current transmission permission.

[0057] The construction process of the distributed time delay complex domain neural network model under the Round-Robin protocol will be described below, and the specific process is as follows:

[0058] (11) Construct a complex-valued neural network model with distributed time delay, which is represented as follows:

[0059]

[0060] where is the component d j is a coefficient matrix with values of j = 1, 2,..., n between 0 and 1, is the state vector of the neuron at time k, is the state vector of the neuron at time k + 1, is the state of the neuron detected by the sensor at time k, is the output vector to be estimated at time k, and are the excitation functions of the neuron at time k and at time k - j, and are the connection weight matrix and the distributed time delay connection weight matrix, respectively, μ(k) is the disturbance input at time k, μ R (k) and μ I (k) represent the real and imaginary components of μ(k), respectively, and satisfy the condition F, C, J, and E are constant matrices, the initial value of the state at time -j is 0, that is, z(-j) = 0 (j = 1, 2,...), and the constant ω j ≥ 0 satisfies the condition and

[0061] Assumption 1: For any real number there exists a positive constant such that the neuron excitation function satisfies the following inequality

[0062]

[0063] where, and are the components of the real and imaginary parts of the neuron excitation function h(·), and are the components of the real and imaginary parts of the neuron excitation function g(·). In addition, at the zero state value, the function takes the value of zero, that is,

[0064] (12) Establish a transmission model for sensor measurement data under the Round-Robin protocol scheduling.

[0065] Consider that the measurement output of the distributed time-delay complex-valued neural network is obtained by m sensors, and only one sensor can obtain the transmission right at each time k under the scheduling of the Round-Robin protocol. Let ρ(k) represent the sensor identifier with the transmission right, which satisfies the conditions ρ(k)∈{1,2,...,m} and ρ(k+m)=ρ(k), and ρ(k)=mod(k-1,m)+1, where mod(k-1,m) represents the remainder after k-1 is divided by m. Let represent the actual transmission measurement of the jth sensor at time k, and the update of the actual transmission measurement of the jth sensor at time k under the scheduling of the Round-Robin protocol is

[0066]

[0067] where λ j (k) is the state of the neuron detected by the jth sensor at time k, and is also the ideal measurement output, and let

[0068]

[0069] be the update matrix, where δ(ρ(k)-j) (j∈{1,2,...,m}) represents the Kronecker function, δ(ρ(k)-j)=1 when ρ(k)=j, and δ(ρ(k)-j)=0 otherwise, and the actual transmission measurement can be described as:

[0070]

[0071] where represents the actual transmission measurement at time k-1, I is the unit matrix, is an augmented vector of m components , that is,

[0072] (13) Establish a distributed time-delay complex-valued neural network model under the Round-Robin protocol.

[0073] First, x(k) and y(k) are used to represent the real part and the imaginary part of the state vector z(k) of the neuron, λ R (k) and λ I (k) represent the real part and the imaginary part of the ideal measurement output λ(k), and represent the real part and the imaginary part of the actual transmission measurement , and represent the real part and the imaginary part of the output to be estimated ​the real and imaginary parts. Augmenting the real and imaginary part vectors, that is, letting

[0074]

[0075] the augmented system comprising the real and imaginary part dynamics of the complex-valued neural network model (1) is obtained and is given by

[0076]

[0077] wherein is the vector augmented with the real part h R (x(k),y(k)) and the imaginary part h I (x(k),y(k)), is the vector augmented with the real part g R (x(k-j),y(k-j)) and the imaginary part g I (x(k-j),y(k-j)), is the vector augmented with the real part μ R (k) and the imaginary part μ I (k). ε are matrices related to the real and imaginary parts of the parameter matrices L, M, F, C, J, E, respectively, and have the form R I R I R I R I R I R I

[0078]

[0079] Since the effect of the zeroth-order hold is considered in the Round-Robin protocol, the augmented vector of the state vector ζ(k) of the neuron at time k and the measurement information actually transmitted at the previous time is denoted by The following distributed time-delay complex-valued neural network model based on the Round-Robin protocol is obtained and is given by

[0080]

[0081] wherein denotes the real part h R (x(k),y(k)) and the imaginary part h I ​​​​​​​​​​​​an augmented vector of (x(k), y(k)), represents the real part g R (x(k-j), y(k-j)) and the imaginary part g I an augmented vector of (x(k), y(k)); is an augmented parameter matrix, which has the following form:

[0082]

[0083] In the embodiment of the present application, the elastic state estimator of the distributed time-delay complex domain neural network model (6) is specifically as follows:

[0084]

[0085] wherein, is the state The state estimation at time k, G ρ(k) is the gain matrix related to the sensor with transmission authority ρ(k), ΔG ρ(k) (k) represents the uncertainty of the gain matrix G ρ(k) , and ΔG ρ(k) (k) = A ρ(k) Γ(k)B ρ(k) wherein, A ρ(k) and B ρ(k) are known real constant matrices, and Γ(k) is a time-varying matrix function satisfying the condition Γ T (k)Γ(k)≤I.

[0086] The state estimation error calculation model (8) is constructed by the distributed time-delay complex domain neural network model (6) and the elastic state estimator (7) under the Round-Robin protocol.

[0087] The state estimation error at time k is defined as Then, according to the distributed time-delay complex domain neural network model (6) and the elastic state estimator (7) based on the Round-Robin protocol, the following state estimation error calculation model (8) can be obtained, which is specifically as follows:

[0088]

[0089] According to the Lyapunov stability theory, the stability of the state estimation error model is analyzed, and when it is detected that the state estimation error model is in a stable state, the gain matrix related to each sensor in the elastic state estimator that satisfies the l2-l ∞ estimation performance index is calculated.

[0090] The state estimation error is considered to be in a stable state and to satisfy the l2-l ∞ The performance index is as follows:

[0091] Condition 1: The state estimation error model (8) is asymptotically stable when the interference input is zero.

[0092] Condition 2: In the case of a distributed time-delay complex neural network model in a zero initial state (the state at time zero is zero) and a non-zero interference input, given an interference attenuation level ψ, the output estimation error satisfies the l2-l ∞ The performance index

[0093] Given an interference attenuation level ψ, if there exists a positive definite matrix P ρ(k) related to the sensor ρ(k) with transmission authority at time k ρ(k+1) related to the sensor ρ(k) with transmission authority at time k ρ(k) , a positive definite matrix Q, diagonal positive definite matrices Θ1 and Θ2, and a positive scalar such that the following inequalities (9), (10) and (11) are satisfied, according to Lyapunov stability theory, the state estimation error calculation model is asymptotically stable and satisfies the given l2-l ∞ The performance index, and the inequalities (9), (10) and (11) are as follows:

[0094]

[0095] wherein, is a constant, and is a parameter matrix of the activation function in the complex neural network model satisfying the above assumption 1, and has the following form:

[0096]

[0097] wherein, Ξ RR , Ξ RI , Ξ IR , Ξ II , Ψ RR , Ψ RI , Ψ IR , Ψ II are parameter matrices.

[0098] Then, the inequalities (9), (10) and (11) are solved by using the linear matrix inequality tool in MATLAB, and the positive definite matrix P ρ(k+1) and the parameter matrix Λ ρ(k) are calculated according to the solution of the linear matrix inequality The gain matrix G ρ(k) of the state estimator can be obtained.

[0099] The above Lyapunov stability theory is:

[0100] V(k+1)-V(k)<0 (13)

[0101]

[0102] Wherein, V(k+1), V(k) are Lyapunov functions at k+1, k time respectively, e T (k) is the transpose of the state estimation error e(k) at k time, is the transpose of , wherein,

[0103] In the embodiment of the application, represents a set of complex n-dimensional vectors, and represent a set of real n×n matrices and n-dimensional vectors respectively. X>Y means that X-Y is a positive definite matrix, wherein X and Y are real symmetric matrices, I represents a unit matrix with appropriate dimensions, 0 represents a zero matrix with appropriate dimensions, * represents an omission symbol of items determined by symmetry in a symmetric block matrix, i represents an imaginary unit, that is ||·|| represents the Euclidean norm, |·| represents the absolute value, l2[0,+∞) represents the square summable space, diag{…} represents the block diagonal matrix, and sup represents the supremum.

[0104] The state estimation method of the complex neural network with distributed time delay described in the application is simulated and verified as follows:

[0105]

[0106] B1=[0.3 0.19 0.4 0.17],

[0107] A1=[0.40.270.20.310.260.20.350.3] T ,B2=[0.150.230.350.2],

[0108] A2=[0.320.30.150.40.30.140.40.26]T .

[0109] Choose the activation function h j (z j,k ) and g j (z j,k )as follows:

[0110]

[0111] According to hypothesis 1,

[0112] In addition, we choose Γ(k) = 0.5cos(k) and ω j =2 -(3+j) (j=1,2,3,…), we can see Setting l2-l ∞ The performance index ψ = 0.6. Solving the linear matrix inequalities (9), (10) and (11), we can obtain the gain matrix related to the two sensors in the elastic state estimator as follows:

[0113]

[0114] Set the initial value to zero and the interference input to μ(k)=sin(k)e -0.1k +sin(k)e -0.1k i, get Figures 2 to 7 The simulation results shown in Figure 2 are as follows: Figure 2 and Figure 3 The state estimation trajectory (curve) and state estimation error trajectory of the neuron under different real state x1(k) and x2(k) are respectively depicted. Figure 4 and Figure 5 The state estimation trajectory and state estimation error trajectory of the neuron in the imaginary state y1(k) and y2(k) are characterized respectively. Figure 6 Characterizes the real part state The state estimation trajectory and state estimation error trajectory under Figure 7 Characterizes the imaginary state The state estimation trajectory and state estimation error trajectory under . Figures 2 to 7 It can be seen that the invented state estimator design method can effectively estimate the target state for complex domain neural networks with distributed time delays under the Round-Robin protocol.

[0115] The present invention has been described exemplarily. Obviously, the specific implementation of the present invention is not limited to the above-mentioned method. As long as various non-substantial improvements are made using the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the scope of protection of the present invention.

Claims

1. A state estimation method for a complex domain neural network with distributed time delay, characterized in that: The method is specifically as follows: (1) Construct a distributed time-delay complex domain neural network model under the Round-Robin protocol and a flexible state estimator for the distributed time-delay complex domain neural network model under the Round-Robin protocol; (2) A state estimation error calculation model is constructed based on a distributed time-delay complex domain neural network model and an elastic state estimator under the Round-Robin protocol; (3) According to Lyapunov stability theory, the stability of the state estimation error model is analyzed. When the state estimation error model is detected to be in a stable state, the calculation satisfies l2-l ∞ The gain matrix associated with each sensor of the elastic state estimator when estimating the performance index; (4) The elastic state estimator receives the state information currently detected by the sensor with transmission authority, and estimates the state of the neurons in the complex domain neural network at the next moment based on the gain matrix related to the sensor with current transmission authority.

2. The state estimation method of a complex domain neural network with distributed time delay according to claim 1, characterized in that: The construction process of the distributed time-delay complex domain neural network model under the Round-Robin protocol is as follows: (11) Constructing a complex domain neural network model with distributed time delay; (12) Establish a transmission model for sensor measurement data under Round-Robin protocol scheduling. The sensor is used to detect the state information of the distributed time-delay complex domain neural network and transmit it to the elastic state estimator; (13) Establish a distributed time-delay complex domain neural network model under the Round-Robin protocol.

3. The state estimation method of a complex domain neural network with distributed time delay according to claim 1, characterized in that: The distributed time-delay complex domain neural network model under the Round-Robin protocol is as follows: in, is the state vector ζ(k) of the neuron at time k and the measurement information actually transmitted at time k-1 The augmented vector, The state vector ζ(k+1) of the neuron at time k+1 and the actual measurement information transmitted at time k The augmented vector, is the real part μ of the interference input μ(k) at time k R (k) and the imaginary part μ I (k) augmented vector, is the augmented parameter matrix, which has the following form:

4. The state estimation method of a complex domain neural network with distributed time delay according to claim 1, characterized in that: The elastic state estimator is as follows: in, is the status The state estimate at time k, G ρ(k) is the gain matrix associated with the sensor ρ(k) with transmission authority, ΔG ρ(k) (k) represents the gain matrix G ρ(k) is uncertain, and ΔG ρ(k) (k) = A ρ(k) Γ(k)B ρ(k) , where A ρ(k) and B ρ(k) is a known real constant matrix, Γ(k) is a matrix that satisfies the condition Γ T (k)Γ(k)≤I time-varying matrix function.

5. The state estimation method of a complex domain neural network with distributed time delay according to claim 1, characterized in that: The state estimation error calculation model is as follows:

6. The state estimation method of a complex domain neural network with distributed time delay according to claim 2, characterized in that: The transmission model of sensor measurement data under Round-Robin protocol scheduling is as follows: The actual transmission measurement of the sensor with transmission permission at time k The details are as follows: Among them, λ j (k) represents the neuron state detected by the jth sensor at time k, let: Where δ(ρ(k)-j)(j∈{1,2,...,m}) represents the Kronecker function. When ρ(k)=j, δ(ρ(k)-j)=1, otherwise δ(ρ(k)-j)=0. The actual transmission measurement The details are as follows: in, represents the measurement actually transmitted at time k-1, and I is the identity matrix.

7. A complex domain neural network elastic state estimation system with distributed time delay under Round-Robin protocol, characterized in that: The system comprises: A distributed time-delay complex domain neural network under the Round-Robin protocol, and m sensors for detecting states of neurons in the distributed time-delay complex domain neural network; An elastic state estimator, communicating with m sensors; A state estimation error calculation model, connected with the elastic state estimator and the distributed time-delay complex domain neural network; a processor connected to the state estimation error calculation model; m sensors detect the current neuron state of the distributed time-delay complex domain neural network. A sensor that obtains the transmission permission outputs the detected current neuron state to the elastic state estimator. The elastic state estimator predicts the state vector of the neuron in the distributed time-delay complex domain neural network at the next moment and outputs it to the state estimation error calculation model. The state estimation error calculation model calculates the state estimation error at the next moment based on the actual state vector of the distributed time-delay complex domain neural network at the next moment. When the state estimation error is stable, the processor calculates the state estimation error that satisfies l2-l ∞ When estimating the performance index, the gain matrix associated with each sensor in the elastic state estimator is used. Then, the elastic state estimator estimates the state of the neurons in the complex domain neural network at the next moment based on the gain matrix associated with the sensor that currently obtains the transmission permission.

Images