Anti-eavesdropping distributed fusion filtering method for multi-rate nonlinear systems
-
2025-12-02
19/006,280
2024-12-31
β Patent granted
US 12,489,785 B1
2025-12-02
-
-
Sakinah White Taylor
Bayramoglu Law Offices LLC
2044-12-31
Smart Summary: A new method helps protect against eavesdropping in sensor networks that monitor complex systems. First, it creates a model to understand how these systems behave over time. Then, it simplifies this model to make calculations easier. The method designs a special filter that can work even when there are outside threats and unreliable data. Finally, it calculates the potential errors in the filtering process to ensure better accuracy and security. π TL;DR
Abstract:
An anti-eavesdropping distributed fusion filtering method for the multi-rate nonlinear system over sensor network includes: Step 1, establishing a dynamic model for the multi-rate nonlinear system over sensor network; Step 2, transforming the multi-rate nonlinear system dynamic model into a single-rate nonlinear system dynamic model through the prediction compensation strategy; Step 3, designing an anti-eavesdropping distributed fusion filter; Step 4, calculating an upper bound on the one-step prediction error covariance (tk+1|tk); Step 5, deriving the local distributed filter parameter Ki(tk+1); Step 6, deriving the selection matrix Lij(tk+1); Step 7, substituting Ki(tk+1) and Lij(tk+1) into Step 3 to obtain the fusion filter {circumflex over (x)}CI(tk+1|+tk+1); Step 8, solving for the upper bound on the local filtering error covariance (tk+1|tk+1). The method solves the problem that the existing fusion filtering method cannot simultaneously deal with the filtering problem for multi-rate nonlinear systems with eavesdroppers and fading measurements, thereby improving the accuracy of the filtering performance.
Inventors:
- Jun Hu 2 π¨π³ Harbin, China
- Hui Yu 2 π¨π³ Harbin, China
- Hongxu Zhang 2 π¨π³ Harbin, China
- Chaoqing Jia 2 π¨π³ Harbin, China
- Zhihui Wu 2 π¨π³ Harbin, China
- Shuting Fan 1 π¨π³ Harbin, China
- Wen Chen 1 π¨π³ Harbin, China
Assignee:
- Harbin University of Science and Technology 3 π¨π³ Harbin, China
Applicant:
Interested in similar patents?
Get notified when new applications in this technology area are published.
Classification:
H04L63/1475 » CPC main
Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic; Countermeasures against malicious traffic Passive attacks, e.g. eavesdropping or listening without modification of the traffic monitored
H04L63/04 » CPC further
Network architectures or network communication protocols for network security for providing a confidential data exchange among entities communicating through data packet networks
H04L9/40 IPC
arrangements for secret or secure communications Cryptographic mechanisms or cryptographic ; Network security protocols Network security protocols
Description
CROSS REFERENCE TO THE RELATED APPLICATIONS
This application is based upon and claims priority to Chinese Patent Application No. 202410423443.5, filed on Apr. 9, 2024, the entire contents of which are incorporated herein by reference.
TECHNICAL FIELD
The present invention relates to a distributed filtering method, in particular to an anti-eavesdropping distributed fusion filtering method for multi-rate nonlinear systems with fading measurements over sensor networks.
BACKGROUND
In recent years, sensor networks have garnered widespread attention for their applications including environmental monitoring, intelligent transportation, industrial control and healthcare, where the filtering algorithm design over sensor networks has emerged as a research focus. Common filtering algorithms include centralized and distributed filtering algorithms. Compared with the former, the distributed filtering algorithm has attracted the research interest of scholars due to its higher reliability, stronger fault tolerance and better scalability. In order to address the limitations of individual sensors, multi-sensor fusion technology has been proposed. Furthermore, systems are often nonlinear in practical engineering, and the sampling rates of different components are generally different because different system components have different physical properties. Therefore, it is of practical significance to investigate the distributed fusion filtering problem for multi-rate nonlinear systems over sensor networks.
Since the communication channel with limited bandwidth cannot accommodate large amounts of data transmission, it may lead to network congestion and fading measurements. In this case, filters may only receive incomplete data, inevitably diminishing the accuracy of filtering algorithms. Thus, it is important to study the influence of fading measurements on the distributed fusion filtering algorithm. On the other hand, considering the openness and sharing nature of sensor networks, the data transmission between sensor nodes is susceptible to eavesdropping. Eavesdroppers could potentially exploit intercepted information to analyze the network topology and then launch broader and more severe cyber-attacks, thereby undermining the security and stability of the network. Hence, effective measures should be implemented to safeguard information security within sensor networks.
The existing distributed fusion filtering methods struggle to concurrently address the filtering issues for multi-rate nonlinear systems with fading measurements and eavesdropping, which can adversely impact filtering performance. Therefore, devising an eavesdropping-resistant distributed filtering approach that accounts for fading measurements holds practical significance.
SUMMARY
In order to address the problem that the existing fusion filtering method cannot simultaneously handle the filtering problem for multi-rate nonlinear systems with eavesdroppers and fading measurements, resulting in low accuracy of filtering performance, the present invention provides an anti-eavesdropping distributed fusion filtering method for multi-rate nonlinear systems.
The present invention adopts the following technical solution.
An anti-eavesdropping distributed fusion filtering method for a multi-rate nonlinear system includes the following steps:
-
- Step 1: establishing a dynamic model for the multi-rate nonlinear system over sensor networks:
x
β‘
(
t
k
+
1
)
=
f
β‘
(
x
β‘
(
t
k
)
)
+
B
β‘
(
t
k
)
β’
Ο
β‘
(
t
k
)
y
i
(
s
k
)
=
Ξ
i
(
s
k
)
β’
C
i
(
s
k
)
β’
x
β‘
(
s
k
)
+
v
i
(
s
k
)
wherein, tk is a state update instant of the multi-rate nonlinear system; x(tk) is a state vector of the multi-rate nonlinear system at time tk; x(tk+1) is a state vector of the multi-rate nonlinear system at time tk+1; f(x(tk)) is a continuous and differentiable nonlinear function with a bounded second-order derivative of the multi-rate nonlinear system at time tk; B(tk) is a coefficient matrix of process noise at time tk; Ο(tk) is process noise with zero mean and covariance Q(tk) at time tk; i is a label of sensor nodes, i=1,2, . . ., N, N represents a number of sensor nodes; sk is a measurement sampling instant of sensor; x(sk) is a state vector of the multi-rate nonlinear system at time sk; yi(sk) is a measurement output signal of an i-th sensor node in the multi-rate nonlinear system at time sk; Ci(sk) is a measurement matrix of the i-th sensor node based on the multi-rate nonlinear system at time sk; vi(sk) is measurement noise of the i-th sensor node in the multi-rate nonlinear system at time sk; Ξi(sk) is used to describe a phenomenon of fading measurements;
-
- Step 2: transforming the dynamic model for the multi-rate nonlinear system over sensor networks in Step 1 into a single-rate nonlinear system dynamic model through a prediction compensation strategy:
y
Β―
i
(
t
k
)
=
(
1
-
Ξ²
β’
(
t
k
)
)
β’
Ξ
Β―
i
β’
C
i
(
t
k
)
β’
x
Λ
i
(
t
k
β
t
k
-
1
)
+
Ξ²
β’
(
t
k
)
β’
y
i
(
t
k
)
Ξ²
β‘
(
t
k
)
=
{
ο ο
1
,
t
k
=
s
t
,
t
=
0
,
1
,
2
,
β¦
ο ο
0
,
otherwise
wherein, yi(tk) is a measurement output signal of the i-th sensor node in a single-rate nonlinear system at time tk; Ξ²(tk) is an auxiliary variable; Ci(tk) is a measurement matrix of the i-th sensor node based on the single-rate nonlinear system at time tk; {circumflex over (x)}i(tk|tkβ1) is a one-step prediction of the i-th sensor node at time tkβ1; Ξi=diag{Ξ»i1,Ξ»i2, . . . , Ξ»iny}; ny is a dimension of yi(sk); Ξ»iu is an expectation of a random variable Ξ»iu(sk), u=1,2, . . . , ny; yi(tk) is a measurement output signal of the i-th sensor node in the multi-rate nonlinear system at time tk; st is the measurement sampling instant of sensor node;
-
- Step 3: designing an anti-eavesdropping distributed fusion filter for the single-rate nonlinear system dynamic model in Step 2; wherein
- Step 3a: when a sensor node exchanges information, in order to prevent transmitted data from being eavesdropped by an eavesdropper and ensure a security of information transmission, adding artificial noise to the one-step prediction {circumflex over (x)}j(tk|tkβ1) of a sensor node j before {circumflex over (x)}j(tk|tkβ1) being sent to the sensor node i:
x
^
ij
o
(
t
k
)
=
L
ij
(
t
k
)
β’
x
^
j
(
t
k
β
t
k
-
1
)
+
(
I
-
L
ij
(
t
k
)
)
β’
a
ij
(
t
k
)
wherein, jβNi, Ni is a set of neighboring nodes of the i-th sensor node; {circumflex over (x)}j(tk|tkβ1) represents a one-step prediction of the j-th sensor node at time
t
k
-
1
;
x
^
ij
o
(
t
k
)
is a transmitted message from the sensor node j to the sensor node i at time tk; I is an nx-dimensional identity matrix; nx is a dimension of the state vector x(tk); aij(tk) is the artificial noise with zero mean and covariance Qij(tk) at time tk; Lij(tk) is a selection matrix at time tk;
-
- Step 3b: when the sensor node i receives the information
x
^
ij
o
β’
(
t
k
)
transmitted by the sensor node j, obtaining a compensated one-step prediction at time tk according to a zero-order holder compensation rule:
x
^
ij
c
(
t
k
)
wherein,
x
^
ij
c
(
t
k
)
=
L
ij
(
t
k
)
β’
x
^
ij
o
(
t
k
)
+
(
I
-
L
ij
(
t
k
)
)
β’
x
^
ij
c
(
t
k
-
1
)
is the compensated one-step prediction at time tk;
x
^
ij
c
(
t
k
-
1
)
is a compensated one-step prediction at time tkβ1;
-
- Step 3c: designing a local distributed filter:
x
Λ
i
(
t
k
+
1
β
t
k
)
=
f
β‘
(
x
Λ
i
(
t
k
β
t
k
)
)
x
Λ
i
(
t
k
+
1
|
t
k
+
1
)
=
x
Λ
i
(
t
k
+
1
|
t
k
)
+
K
i
(
t
k
+
1
)
β’
(
y
Β―
i
(
t
k
+
1
)
-
Ξ
Β―
i
β’
C
i
(
t
k
+
1
)
β’
x
Λ
i
(
t
k
+
1
|
t
k
)
)
+
Ξ΅
i
β’
β
j
β
N
i
β’
h
i
β’
j
(
x
Λ
i
(
t
k
+
1
|
t
k
)
-
x
Λ
i
β’
j
c
(
t
k
+
1
)
)
wherein, {circumflex over (x)}i(tk+1|tk) represents a one-step prediction of the i-th sensor node at time tk;
x
^
ij
c
(
t
k
+
1
)
represents a compensated one-step prediction at time tk+1; {circumflex over (x)}i(tk+1|tk+1) represents a filter of the i-th sensor node at time tk+1; {circumflex over (x)}i(tk|tk) represents a filter of the i-th sensor node at time tk; f({circumflex over (x)}i(tk|tk)) represents a nonlinear function filtering form based on the single-rate nonlinear system of the i-th sensor node at time tk; Ki(tk+1) represents a local distributed filter parameter of the i-th sensor node at time tk+1; yi(tk+1) represents a measurement output signal of the i-th sensor node in the single-rate nonlinear system at time tk+1; Ci(tk+1) is a measurement matrix of the i-th sensor node based on the single-rate nonlinear system at time tk+1; Ξ΅i represents a predefined consensus parameter of the i-th sensor node; hij represents a connection coefficient between the i-th sensor node and the j-th sensor node; and
-
- Step 3d: obtaining an anti-eavesdropping distributed fusion filter based on a local filter {circumflex over (x)}i(tk|tk) and a covariance intersection fusion criterion:
C
β’
I
(
t
k
|
t
k
)
=
(
β
t
=
1
N
β’
Ο
t
i
-
1
(
t
k
|
t
k
)
)
-
1
x
Λ
C
β’
I
(
t
k
|
t
k
)
=
C
β’
I
(
t
k
|
t
k
)
β’
β
i
=
1
N
β’
Ο
i
t
-
1
(
t
k
|
t
k
)
β’
x
Λ
t
(
t
k
|
t
k
)
wherein, a superscript ββ1β represents an inverse of a matrix; {circumflex over (x)}CI(tk|tk) is a fusion filter at time tk; CI(tk|tk) is fusion filtering error covariance at time tk; i(tk|tk) is an upper bound on a local filtering error covariance of the i-th sensor node at time tk; iβ1(tk|tk) is an inverse of a matrix
β΅
i
(
t
k
β
t
k
)
;
(
β
i
=
1
N
Ο
i
β’
β΅
i
-
1
(
t
k
β
t
k
)
)
-
1
is an inverse of a matrix
β
i
=
1
N
Ο
i
β’
β΅
i
-
1
(
t
k
β
t
k
)
;
Οi is a scalar;
-
- Step 4: calculating an upper bound on the one-step prediction error covariance i(tk+1|tk) of the i-th sensor node at time tk by solving a matrix difference equation:
β΅
i
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
=
(
1
+
o
1
`
)
β’
A
i
(
t
k
)
β’
β΅
i
(
t
k
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
A
i
T
(
t
k
)
+
B
β‘
(
t
k
)
β’
Q
β‘
(
t
k
)
β’
B
T
(
t
k
)
+
(
1
+
o
`
1
-
1
)
β’
tr
β’
{
M
i
(
t
k
)
β’
β΅
i
(
t
k
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
M
i
T
(
t
k
)
}
β’
D
i
(
t
k
)
β’
D
i
T
(
t
k
)
wherein, a superscript βTβ represents the transpose of a matrix; Γ²1 is a known scaling parameter; Γ²1β1 is an inverse of Γ²1; Ai(tk) is a partial derivative of the continuous and differentiable nonlinear function f(x(tk)) corresponding to a system state at the local filter {circumflex over (x)}i(tk|tk) at time tk; Mi(tk) and Di(tk) are known error matrices obtained by Taylor series based on f(x(tk));
A
i
T
(
t
k
)
,
BT(tk),
M
i
T
(
t
k
)
and
D
i
T
(
t
k
)
represent transposes of Ai(tk), B(tk), Mi(tk) and Di(tk), respectively;
-
- Step 5: according to (tk+1|tk) obtained in Step 4, deriving the local distributed filter parameter Ki(tk+1) of the i-th sensor node at time tk+1 by minimizing a trace of the upper bound on the local filtering error covariance:
Ki(tk+1)=Ξ΄(Ξ²(tk+1),1)(1+Γ²2)(tk+1|tk)CiT(tk+1)ΞiΞ iβ1(tk+1)
wherein,
- Step 5: according to (tk+1|tk) obtained in Step 4, deriving the local distributed filter parameter Ki(tk+1) of the i-th sensor node at time tk+1 by minimizing a trace of the upper bound on the local filtering error covariance:
β
j
(
t
k
+
1
)
=
(
1
+
o
2
`
)
β’
Ξ
Β―
i
β’
C
i
(
t
k
+
1
)
β’
β΅
i
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
C
i
T
(
t
k
+
1
)
β’
Ξ
Β―
i
+
tr
β’
{
C
i
(
t
k
+
1
)
β’
X
_
i
(
t
k
+
1
)
β’
C
i
T
(
t
k
+
1
)
}
β’
Ξ
Λ
i
+
R
i
(
t
k
+
1
)
β’
X
_
i
(
t
k
+
1
)
=
(
1
+
o
`
3
-
1
)
β’
x
Λ
i
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
x
Λ
i
T
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
+
(
1
+
o
3
`
)
β’
β΅
i
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
wherein, Ξ΄(a,b) is a Kronecker function; Ξ²(tk+1) is an auxiliary variable at time tk+1; Γ²2 and Γ²3 are known scaling parameters; Γ²3β1 is an inverse of Γ²3; CiT(tk+1) is a transpose of Ci(tk+1);
x
Λ
i
T
(
t
k
+
1
|
t
k
)
is a transpose of {circumflex over (x)}i(tk+1|tk); Ξ iβ1(tk+1) is an inverse of Ξ i(tk+1); Ri(tk+1) is a covariance matrix of measurement noise vi(tk+1) of the i-th sensor node at time tk+1;
-
- Step 6: by maximizing an estimation error covariance of the eavesdropper, deriving the selection matrix Lij(tk+1) at time tk+1 from the following optimization problem:
max
L
ij
(
t
k
+
1
)
β
L
^
ij
(
t
k
+
1
)
t
β’
r
β’
{
W
ij
(
t
k
+
1
)
}
wherein,
W
i
β’
j
(
t
k
+
1
)
=
(
I
-
L
i
β’
j
(
t
k
+
1
)
)
β’
(
2
β’
x
Λ
j
(
t
k
+
1
|
t
k
)
β’
x
Λ
j
T
(
t
k
+
1
|
t
k
)
+
Q
ij
(
t
k
+
1
)
)
β’
(
I
-
L
ij
(
t
k
+
1
)
)
wherein,
L
Λ
ij
(
t
k
+
1
)
=
{
L
i
β’
j
1
(
t
k
+
1
)
,
L
ij
2
(
t
k
+
1
)
,
β¦
,
L
ij
d
(
t
k
+
1
)
}
,
L
ij
1
(
t
k
+
1
)
,
L
ij
2
(
t
k
+
1
)
,
β¦
,
L
ij
d
i
(
t
k
+
1
)
are diagonal matrices with elements of 0 or 1 and a sum of the diagonal elements are Γ±i; {circumflex over (x)}j(tk+1|tk) represents a one-step prediction of the j-th sensor node at time
t
k
;
x
Λ
j
T
(
t
k
+
1
|
t
k
)
is a transpose of {circumflex over (x)}j(tk+1|tk); Qij(tk+1) is a covariance matrix of artificial noise aij(tk+1) at time tk+1;
max
x
f
β‘
(
x
)
represents that an objective function f(x) is maximized by selecting a decision variable x;
-
- Step 7: substituting Ki(tk+1) obtained in Step 5 and Lij(tk+1) obtained in Step 6 into Step 3 to obtain the fusion filter {circumflex over (x)}CI(tk+1|tk+1) at time tk+1; determining whether tk+1 reaches a total duration M, if tk+1<M, performing Step 8, otherwise, ending;
- Step 8: based on Ki(tk+1) obtained in Step 5 and Lij(tk+1) obtained in Step 6, solving for the upper bound on the local filtering error covariance (tk+1|tk+1) of the i-th sensor node at time tk+1:
(tk+1|tk+1)=(1βΞ²(tk+1))Ξi(tk+1)+Ξ²(tk+1)Ξi(tk+1)
wherein,
Ξ
i
(
t
k
+
1
)
=
(
1
+
Γ²
2
)
β’
β΅
i
(
t
k
+
1
β
t
k
)
+
(
1
+
Γ²
2
-
1
)
β’
Ξ΅
i
2
β’
ΓΈ
i
β’
β
j
β
N
i
h
ij
β’
X
ij
(
t
k
+
1
)
β’
X
ij
T
(
t
k
+
1
)
β’
Ξ
i
(
t
k
+
1
)
=
(
1
+
Γ²
2
)
β’
(
I
-
K
i
(
t
k
+
1
)
β’
Ξ
_
i
β’
C
i
(
t
k
+
1
)
)
β’
β΅
i
(
t
k
+
1
β
t
k
)
β’
(
I
-
K
i
(
t
k
+
1
)
β’
Ξ
_
i
β’
C
i
(
t
k
+
1
)
)
T
+
(
1
+
Γ²
2
-
1
)
β’
Ξ΅
i
2
β’
ΓΈ
i
β’
β
j
β
N
i
h
ij
β’
X
ij
(
t
k
+
1
)
β’
X
ij
T
(
t
k
+
1
)
β’
K
i
(
t
k
+
1
)
β’
R
i
(
t
k
+
1
)
β’
K
i
T
(
t
k
+
1
)
+
tr
β’
{
C
i
(
t
k
+
1
)
β’
X
_
i
(
t
k
+
1
)
β’
C
i
T
(
t
k
+
1
)
}
β’
Ξ
_
i
β’
K
i
(
t
k
+
1
)
β’
K
i
T
(
t
k
+
1
)
β’
X
ij
(
t
k
+
1
)
=
L
ij
(
t
k
+
1
)
β’
x
~
ij
(
t
k
+
1
)
+
(
I
-
L
ij
(
t
k
+
1
)
)
β’
x
~
ij
c
(
t
k
+
1
)
β’
x
~
ij
c
(
t
k
+
1
)
=
x
^
i
(
t
k
+
1
β
t
k
)
-
x
^
ij
c
(
t
k
)
,
x
~
ij
(
t
k
+
1
)
=
x
^
i
(
t
k
+
1
β
t
k
)
-
x
^
j
(
t
k
+
1
β
t
k
)
wherein, Γ²2β1 is an inverse of Γ²2; Ξ΅i2 is a square of Ξ΅i; (tk+1|tk+1) is the upper bound on the local filtering error covariance of the i-th sensor node at time tk+1;
K
i
T
(
t
k
+
1
)
is a transpose of Ki(tk+1);
X
ij
T
(
t
k
+
1
)
is a transpose of Xij(tk+1); (IβKi(tk+1)ΞiCi(tk+1))T is a transpose of IβKi(tk+1)ΞiCo(tk+1); ΓΈi represents a penetration of the i-th sensor node;
let tk=tk+1 and performing Step 3 until tk+1=M is satisfied.
Compared with the prior art, the present invention has the following advantages.
The present invention simultaneously takes into account the influence of eavesdroppers and fading measurements on the filtering performance, obtaining the minimum upper bound on the local filtering error covariance based on the Riccati differential equation. In particular, during the information transmission between sensor nodes, artificial noise is introduced to achieve the purpose of eavesdropping resistance.
The present invention employs a recursive approach to estimate the states of multi-rate nonlinear systems, which is advantageous for its solvability and suitability for online implementation. At the same time, such fusion filtering problems can be categorized into centralized and distributed fusion filtering issues. Notably, in the centralized fusion filtering method, the measurement outputs from all sensor nodes must be expressed in a compact form, which incurs high computational costs for sensor networks with large nodes. In contrast, the distributed fusion filtering method utilized in the present invention has a lower computational burden.
The present invention addresses the limitations of existing fusion filtering methods in handling multi-rate nonlinear systems that are subject to eavesdroppers and fading measurements, thereby enhancing the accuracy of filtering performance for such issues. It can be seen from the simulation figures that the average mean square error of the fusion filter decreases as the increases of fading probability Ξ»iu. Specifically, when a fading probability rises from 0.3 to 0.7, an average mean square error is reduced by approximately 37%; and when it increases from 0.7 to 0.8, a reduction is approximately 88%.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 shows a diagram of the anti-eavesdropping distributed fusion filtering algorithm.
FIG. 2 shows trajectories of the first component of the actual state x1(tk) and its fusion filter {circumflex over (x)}CL1(tk|tk) for the multi-rate nonlinear system.
FIG. 3 shows trajectories of the second component of the actual state x2(tk) and its fusion filter {circumflex over (x)}CL2(tk|tk) for the multi-rate nonlinear system.
FIG. 4 shows the logarithm of the mean square error log(MSEi(tk)) and the logarithm of the trace of the minimum upper bound on the local filtering error covariance log(tr{(tk|tk)}) for four sensor nodes.
FIG. 5 shows the cumulative error square for four sensor nodes and the fusion filter.
FIG. 6 shows log(tr{i(tk|tk)}) under different fading probabilities for four sensor nodes.
FIG. 7 shows log(tr{CI(tk|tk)}) under different fading probabilities for the fusion filter.
FIG. 8 shows log(MSECI(tk)) under different fading probabilities for the fusion filter.
DETAILED DESCRIPTION OF THE EMBODIMENTS
The embodiment of the present invention is further described in detail hereinafter in conjunction with the figures.
The present invention provides an anti-eavesdropping distributed fusion filtering algorithm, as shown in FIG. 1, the present invention includes the following detailed steps:
-
- Step 1: establishing a dynamic model for the multi-rate nonlinear system over sensor networks;
in this step, establishing a dynamic model for the multi-rate nonlinear system over sensor networks:
- Step 1: establishing a dynamic model for the multi-rate nonlinear system over sensor networks;
x
β‘
(
t
k
+
1
)
=
f
β‘
(
x
β‘
(
t
k
)
)
+
B
β‘
(
t
k
)
β’
Ο
β‘
(
t
k
)
β’
y
i
(
s
k
)
=
Ξ
i
(
s
k
)
β’
C
i
(
s
k
)
β’
x
β‘
(
s
k
)
+
v
i
(
s
k
)
(
1
)
wherein, tk is a state update instant of the multi-rate nonlinear system, any two adjacent state update instants tk and tk+1 satisfy tk+1βtk=h, h>0 is the state update period of the multi-rate nonlinear system; x(tk) is a state vector of the multi-rate nonlinear system at time tk; x(tk+1) is a state vector of the multi-rate nonlinear system at time tk+1; f(x(tk)) is a continuous and differentiable nonlinear function with a bounded second-order derivative of the multi-rate nonlinear system at time tk; B(tk) is a coefficient matrix of process noise at time tk; Ο(tk) is process noise with zero mean and covariance Q(tk) at time tk; i is a label of sensor nodes, i=1,2, . . . , N, N represents a number of sensor nodes; sk is a measurement sampling instant of sensor, any two adjacent measurement sampling instants sk and sk+1 satisfy sk+1βsk=bh, bh is a measurement sampling period of sensor node, b is a positive integer; x(sk) is a state vector of the multi-rate nonlinear system at time sk; yi(sk) is a measurement output signal of an i-th sensor node in the multi-rate nonlinear system at time sk; Ci(sk) is a measurement matrix of the i-th sensor node based on the multi-rate nonlinear system at time sk; vi(sk) is measurement noise of the i-th sensor node in the multi-rate nonlinear system at time sk with zero mean and covariance Ri(sk); Ξi(sk)=diag {Ξ»i1(sk),Ξ»i2(sk), . . . , Ξ»iny(sk)} is used to describe a phenomenon of fading measurements, diag{β‘} represents a diagonal matrix composed of elements ββ; ny is a dimension of the measurement output vector yi(sk); the elements Ξ»iu(sk) (u=1,2, . . . , ny) are random variables distributed over an interval [0,1], which meet E{Ξ»iu(sk)}=Ξ»iu and
E
β’
{
(
Ξ»
iu
(
s
k
)
-
Ξ»
_
iu
)
2
}
=
E
β’
{
Ξ»
~
i
β’
u
2
(
s
k
)
}
=
Ξ»
Λ
i
β’
u
,
Ξ»iu and {circumflex over (Ξ»)}iu are expectation and variance of the random variable Ξ»iu(sk), respectively, E{} is a mathematical expectation;
-
- Step 2: transforming the multi-rate nonlinear system dynamic model over sensor networks in Step 1 into a single-rate nonlinear system dynamic model through a prediction compensation strategy: wherein detailed steps are as follows:
for the sensor node i, defining the auxiliary variable Ξ²(tk):
- Step 2: transforming the multi-rate nonlinear system dynamic model over sensor networks in Step 1 into a single-rate nonlinear system dynamic model through a prediction compensation strategy: wherein detailed steps are as follows:
Ξ²
β‘
(
t
k
)
=
{
1
,
t
k
=
s
t
,
t
=
0
,
1
,
2
,
β¦
0
,
otherwise
obtaining the measurement output model of the single-rate nonlinear system as follows:
yi(tk)=(1βΞ²(tk))ΞiCi(tk){circumflex over (x)}i(tk|tkβ1)+Ξ²(tk)yi(tk)ββ (2)
wherein, st (t=0,1,2, . . . ) is the measurement sampling instant of sensor node; yi(tk) is a measurement output signal of the i-th sensor node in a single-rate nonlinear system at time tk; Ci(tk) is a measurement matrix of the i-th sensor node based on the single-rate nonlinear system at time tk; {circumflex over (x)}i(tk|tkβ1) is a one-step prediction of the i-th sensor node at time tkβ1; Ξi=diag{Ξ»i1,Ξ»i2, . . . , Ξ»iny}; yi(tk) is a measurement output signal of the i-th sensor node in the multi-rate nonlinear system at time tk;
-
- Step 3: designing an anti-eavesdropping distributed fusion filter for the single-rate nonlinear system dynamic model in Step 2; wherein detailed steps are as follows:
- Step 3a: when a sensor node exchanges information, in order to prevent transmitted data from being eavesdropped by an eavesdropper and ensure a security of information transmission, adding artificial noise to the one-step prediction {circumflex over (x)}j(tk|tkβ1) of a sensor node j (jβNi) before {circumflex over (x)}j(tk|tkβ1) being sent to the sensor node i:
x
^
ij
o
(
t
k
)
=
L
ij
(
t
k
)
β’
x
Λ
j
(
t
k
|
t
k
-
1
)
+
(
I
-
L
ij
(
t
k
)
)
β’
a
ij
(
t
k
)
(
3
)
wherein, Ni is a set of neighboring nodes of the i-th sensor node; {circumflex over (x)}j(tk|tkβ1) is a one-step prediction of the j-th sensor node at time tkβ1;
x
^
ij
o
(
t
k
)
is a transmitted message from the sensor node j to the sensor node i at time tkl I is an nx-dimensional identity matrix, nx is a dimension of the state vector x(tk); aij(tk) is the artificial noise with zero mean and covariance Qij(tk) at time tk;
L
ij
(
t
k
)
=
diag
(
β
ij
1
(
t
k
)
,
β
ij
2
(
t
k
)
,
β¦
,
β
ij
n
x
(
t
k
)
}
is a selection matrix at time tk, elements
β
ij
1
(
t
k
)
,
β
ij
2
(
t
k
)
,
β¦
,
β
ij
n
x
(
t
k
)
satisfy
β
ij
m
(
t
k
)
β
{
0
,
1
}
β’
(
m
=
1
,
2
,
β¦
,
n
X
)
and
β
m
=
1
n
x
β
ij
m
(
t
k
)
=
n
~
i
,
n
~
i
is a constant that is pre-designed according to an actual demand; βΞ£β is a summation symbol. In addition, Lij(tk)β{circumflex over (L)}ij(tk), wherein,
L Λ i β’ j ( t k ) = { L i β’ j 1 ( t k ) , L i β’ j 2 ( t k ) , β¦ , L ij d ( t k ) } , L ij 1 ( t k ) , L ij 2 ( t k ) , β¦ ,
L
ij
d
i
(
t
k
)
are all diagonal matrices with elements of 0 or 1 and the sum of the diagonal elements is Γ±,
d
i
=
C
n
x
n
~
i
,
C
n
x
n
~
i
represents a number of combinations;
-
- Step 3b: when the sensor node i receives the information
x
Λ
ij
o
(
t
k
)
transmitted by the sensor node j, obtaining a compensated one-step prediction at time tk according to a zero-order holder compensation rule:
x
Λ
i
β’
j
c
(
t
k
)
=
L
i
β’
j
(
t
k
)
β’
x
Λ
i
β’
j
o
(
t
k
)
+
(
I
-
L
ij
(
t
k
)
)
β’
x
Λ
i
β’
j
c
(
t
k
-
1
)
(
4
)
wherein,
x
Λ
ij
c
(
t
k
)
is the compensated one-step prediction at time tk,
x
Λ
ij
c
(
t
k
-
1
)
is a compensated one-step prediction at time tkβ1;
-
- Step 3c: designing a local distributed filter:
{circumflex over (x)}i(tk+1|tk)=f(x, ({circumflex over (x)}j(tk|tk))ββ(5)
- Step 3c: designing a local distributed filter:
x
^
i
(
t
k
+
1
β
t
k
+
1
)
=
x
^
i
(
t
k
+
1
β
t
k
)
+
K
i
(
t
k
+
1
)
β’
(
y
_
i
(
t
k
+
1
)
-
Ξ
_
i
β’
C
i
(
t
k
+
1
)
β’
x
^
i
(
t
k
+
1
β
t
k
)
)
+
Ξ΅
i
β’
β
j
β
N
i
h
ij
(
x
^
i
(
t
k
+
1
β
t
k
)
-
x
^
ij
c
(
t
k
+
1
)
)
(
6
)
wherein, {circumflex over (x)}i(tk+1|tk) represents a one-step prediction of the i-th sensor node at time tk;
x
Λ
ij
c
(
t
k
+
1
)
represents a compensated one-step prediction at time tk+1; {circumflex over (x)}i(tk+1|tk+1) represents a filter of the i-th sensor node at time tk+1; {circumflex over (x)}i(tk|tk) represents a filter of the i-th sensor node at time tk; f({circumflex over (x)}i(tk|tk)) represents a nonlinear function filtering form based on the single-rate nonlinear system of the i-th sensor node at time tk; Ki(tk+1) represents a local distributed filter parameter of the i-th sensor node at time tk+1; yi(tk+1) represents a measurement output signal of the i-th sensor node in the single-rate nonlinear system at time tk+1; Ci(tk+1) is a measurement matrix of the i-th sensor node based on the single-rate nonlinear system at time tk+1; Ξ΅i represents a predefined consensus parameter of the i-th sensor node; hij represents a connection coefficient between the i-th sensor node and the j-th sensor node; and
-
- Step 3d: obtaining the distributed fusion filter based on a local filter {circumflex over (x)}i(tk|tk) (i=1,2, . . . , N) and a covariance intersection fusion criterion:
β΅
CI
(
t
k
β
t
k
)
=
(
β
i
=
1
N
Ο
i
β’
β΅
i
-
1
(
t
k
β
t
k
)
)
-
1
β’
x
^
CI
(
t
k
β
t
k
)
=
β΅
CI
(
t
k
β
t
k
)
β’
β
i
=
1
N
Ο
i
β’
β΅
i
-
1
(
t
k
β
t
k
)
β’
x
^
i
(
t
k
β
t
k
)
(
7
)
wherein, a superscript ββ1β represents an inverse of a matrix; {circumflex over (x)}CI(tk|tk) is a fusion filter at time tk; CI(tk|tk) is fusion filtering error covariance at time tk; i(tk|tk) is an upper bound on a local filtering error covariance of the i-th sensor node at time tk; iβ1(tk|tk) is an inverse of a matrix
β΅
i
(
t
k
β
t
k
)
;
(
β
i
=
1
N
Ο
i
β’
β΅
i
-
1
(
t
k
β
t
k
)
)
-
1
is an inverse of a matrix
β
i
=
1
N
Ο
i
β’
β΅
i
-
1
(
t
k
β
t
k
)
;
Οi is a scalar, which can be obtained by a following optimization problem:
min
Ο
i
tr
β’
{
β΅
CI
(
t
k
β
t
k
)
}
β’
s
.
t
.
β
i
=
1
N
Ο
i
=
1
,
Ο
i
β₯
0
(
8
)
wherein, tr{β‘} is a trace of a corresponding matrix;
min
x
f
β‘
(
x
)
means that the objective function f(x) is minimized by selecting the decision variable x, s.t. is an abbreviation for βsubject toβ;
-
- Step 4: calculating an upper bound on the one-step prediction error covariance (tk+1|tk) of the i-th sensor node at time tk by solving a matrix difference equation;
in this step, the upper bound on the one-step prediction error covariance (tk+1|tk) is calculated according to the following equation:
- Step 4: calculating an upper bound on the one-step prediction error covariance (tk+1|tk) of the i-th sensor node at time tk by solving a matrix difference equation;
β΅
i
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
=
(
1
+
o
1
`
)
β’
A
i
(
t
k
)
β’
β΅
i
(
t
k
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
A
i
T
(
t
k
)
+
B
β‘
(
t
k
)
β’
Q
β‘
(
t
k
)
β’
B
T
(
t
k
)
+
(
1
+
o
`
1
-
1
)
β’
tr
β’
{
M
i
(
t
k
)
β’
β΅
i
(
t
k
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
M
i
T
(
t
k
)
}
β’
D
i
(
t
k
)
β’
D
i
T
(
t
k
)
(
9
)
wherein, a superscript βTβ represents a transpose of the matrix; Γ²1 is a known scaling parameter; Γ²1β1 is an inverse of {grave over (p)}1; Ai(tk) is a partial derivative of the continuous and differentiable nonlinear function f(x tk)) corresponding to a system state at the local filter {circumflex over (x)}i(tk|tk) at time tk; Mi(tk) and Di(tk) are known error matrices obtained by Taylor series based on f(x(tk));
A
i
T
(
t
k
)
,
BT(tk),
M
i
T
(
t
k
)
β’
and
β’
D
i
T
(
t
k
)
represent transposes of Ai(tk), B(tk), Mi(tk) and Di(tk), respectively;
-
- Step 5: According to (tk+1|tk) obtained in Step 4, deriving the local distributed filter parameter Ki(tk+1) of the i-th sensor node at time tk+1;
in this step, the local distributed filter parameter Ki(tk+1) is derived by minimizing a trace of the upper bound on the local filtering error covariance:
Ki(tk+1)=Ξ΄(Ξ²(tk+1),1)(1+Γ²2)i(tk+1|tk)CiT(tk+1)ΞiΞ iβ1(tk+1)ββ(10)
wherein,
- Step 5: According to (tk+1|tk) obtained in Step 4, deriving the local distributed filter parameter Ki(tk+1) of the i-th sensor node at time tk+1;
β
j
(
t
k
+
1
)
=
(
1
+
o
2
`
)
β’
Ξ
Β―
i
β’
C
i
(
t
k
+
1
)
β’
β΅
i
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
C
i
T
(
t
k
+
1
)
β’
Ξ
Β―
i
+
tr
β’
{
C
i
(
t
k
+
1
)
β’
X
_
i
(
t
k
+
1
)
β’
C
i
T
(
t
k
+
1
)
}
β’
Ξ
Λ
i
+
R
i
(
t
k
+
1
)
β’
X
_
i
(
t
k
+
1
)
=
(
1
+
o
`
3
-
1
)
β’
x
Λ
i
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
x
Λ
i
T
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
+
(
1
+
o
3
`
)
β’
β΅
i
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
wherein, Ξ΄(a,b) is a Kronecker function and meets
Ξ΄
β‘
(
a
,
b
)
=
β’
{
1
,
a
=
b
0
,
a
β
b
;
Ξ²
β‘
(
t
k
+
1
)
is an auxiliary variable at time tk+1; Γ²2 and Γ²3 are known scaling parameters; Γ²3β1 is an inverse of Γ²3;
C
i
T
(
t
k
+
1
)
is a transpose of Ci(tk+1); {circumflex over (Ξ)}i=diag{({circumflex over (Ξ»)}i1,{circumflex over (Ξ»)}i2, . . . , {circumflex over (Ξ»)}iny}; {circumflex over (x)}iT(tk+1|tk) is a transpose of {circumflex over (x)}i(tk+1|tk); Ξ iβ1(tk+1) is an inverse of Ξ i(tk+1); Ri(tk+1) is a covariance matrix of measurement noise vi(tk+1) of the i-th sensor node at time tk+1;
-
- Step 6: by maximizing an estimation error covariance of the eavesdropper, deriving the selection matrix Lij(tk+1) at time tk+1;
in this step, obtaining the selection matrix Lij(tk+1) from the following optimization problem:
- Step 6: by maximizing an estimation error covariance of the eavesdropper, deriving the selection matrix Lij(tk+1) at time tk+1;
max
L
ij
(
t
k
+
1
)
β
L
^
ij
(
t
k
+
1
)
t
β’
r
β’
{
W
ij
(
t
k
+
1
)
}
(
11
)
wherein,
W
i
β’
j
(
t
k
+
1
)
=
(
I
-
L
i
β’
j
(
t
k
+
1
)
)
β’
(
2
β’
x
Λ
j
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
β’
x
Λ
j
T
(
t
k
+
1
β’
β
"\[LeftBracketingBar]"
t
k
)
+
Q
i
β’
j
(
t
k
+
1
)
)
β’
(
I
-
L
i
β’
j
(
t
k
+
1
)
)
wherein,
L
Λ
ij
(
t
k
+
1
)
=
{
L
i
β’
j
1
(
t
k
+
1
)
,
L
ij
2
(
t
k
+
1
)
,
β¦
,
L
ij
d
(
t
k
+
1
)
}
,
L
ij
1
(
t
k
+
1
)
,
L
ij
2
(
t
k
+
1
)
,
β¦
,
L
ij
d
i
(
t
k
+
1
)
are diagonal matrices with elements of 0 or 1 and a sum of the diagonal elements are Γ±i; {circumflex over (x)}j(tk+1|tk) represents a one-step prediction of the j-th sensor node at time tk;
x
Λ
j
T
(
t
k
+
1
|
t
k
)
is a transpose of {circumflex over (x)}i(tk+1|tk); Qij(tk+1) is a covariance matrix of artificial noise aij(tk+1) at time tk+1;
max
x
f
β‘
(
x
)
represents that an objective function f(x) is maximized by selecting a decision variable x;
-
- Step 7: substituting Ki(tk+1) obtained in Step 5 and Lij(tk+1) obtained in Step 6 into Step 3 to obtain the fusion filter {circumflex over (x)}CI(tk+1|tk+1) at time tk+1; determining whether tk+1 reaches a total duration M, if tk+1<M, performing Step 8, otherwise, ending;
- Step 8: based on Ki(tk+1) obtained in Step 5 and Lij(tk+1) obtained in Step 6, solving for the upper bound on the local filtering error covariance (tk+1|tk+1) of the i-th sensor node at time tk+1; let tk=tk+1 and performing Step 3 until tk+1=M is satisfied;
in this step, the upper bound on the local filtering error covariance (tk+1|tk+1) is calculated as follows:
(tk+1|tk+1)=(1=Ξ²(tkβ1))Ξi(tk+1)+Ξ²(tk+1)Ξi(tk+1)ββ(12)
wherein,
Ξ
i
(
t
k
+
1
)
=
(
1
+
Γ²
2
)
β’
β΅
i
(
t
k
+
1
β
t
k
)
+
(
1
+
Γ²
2
-
1
)
β’
Ξ΅
i
2
β’
ΓΈ
i
β’
β
j
β
N
i
h
ij
β’
X
ij
(
t
k
+
1
)
β’
X
ij
T
(
t
k
+
1
)
β’
Ξ
i
(
t
k
+
1
)
=
(
1
+
Γ²
2
)
β’
(
I
-
K
i
(
t
k
+
1
)
β’
Ξ
_
i
β’
C
i
(
t
k
+
1
)
)
β’
β΅
i
(
t
k
+
1
β
t
k
)
β’
(
I
-
K
i
(
t
k
+
1
)
β’
Ξ
_
i
β’
C
i
(
t
k
+
1
)
)
T
+
(
1
+
Γ²
2
-
1
)
β’
Ξ΅
i
2
β’
ΓΈ
i
β’
β
j
β
N
i
h
ij
β’
X
ij
(
t
k
+
1
)
β’
X
ij
T
(
t
k
+
1
)
β’
K
i
(
t
k
+
1
)
β’
R
i
(
t
k
+
1
)
β’
K
i
T
(
t
k
+
1
)
+
tr
β’
{
C
i
(
t
k
+
1
)
β’
X
_
i
(
t
k
+
1
)
β’
C
i
T
(
t
k
+
1
)
}
β’
Ξ
_
i
β’
K
i
(
t
k
+
1
)
β’
K
i
T
(
t
k
+
1
)
β’
X
ij
(
t
k
+
1
)
=
L
ij
(
t
k
+
1
)
β’
x
~
ij
(
t
k
+
1
)
+
(
I
-
L
ij
(
t
k
+
1
)
)
β’
x
~
ij
c
(
t
k
+
1
)
β’
x
~
ij
c
(
t
k
+
1
)
=
x
^
i
(
t
k
+
1
β
t
k
)
-
x
^
ij
c
(
t
k
)
,
x
~
ij
(
t
k
+
1
)
=
x
^
i
(
t
k
+
1
β
t
k
)
-
x
^
j
(
t
k
+
1
β
t
k
)
wherein, Γ²2β1 is an inverse of Γ²2;
Ξ΅
i
2
is a square of Ξ΅i; (tk+1|tk+1) is the upper bound on the local filtering error covariance of the i-th sensor node at time tk+1;
K
i
T
(
t
k
+
1
)
is a transpose of Ki(tk+1);
X
ij
T
(
t
k
+
1
)
is a transpose of Xij(tk+1); (IβKi(tk+1)ΞiCi(tk+1))T is a transpose of IβKi(tk+1)ΞiCi(tk+1); ΓΈi represents a penetration of the i-th sensor node;
in this step, (tk+1|tk+1) is calculated for each sensor node, such that Pi(tk+1|tk+1)β€(tk+1|tk+1) holds, wherein Pi(tk+1|tk+1) is the local filtering error covariance of the i-th sensor node at time tk+1. Furthermore, the local distributed filter parameter Ki(tk+1) is designed by minimizing the trace of (tk+1|tk+1) at time tk+1.
Embodiment: the distributed fusion filtering problem for multi-rate nonlinear systems subject to eavesdroppers and fading measurements is solved by adopting the fusion filtering method proposed in the present invention.
In the present invention, the sensor network with four sensor nodes is selected for simulation experiment with the following edge set:
E={(1,3),(1,4),(2,1),(2,4),(3,1),(3,2),(4,1),(4,3)}
wherein, each ordinal pair represents the interaction of information between sensor nodes, for example, (1,3) means that the third sensor node can transmit information to the first sensor node. For (i, j) (i, j=1,2,3,4), if (i, j)βE, then the adjacency coefficient between the i-th sensor node and the j-th sensor node is set as hij=1, otherwise hij=0.
The system matrices are set as follows:
f
β‘
(
x
β‘
(
t
k
)
)
=
[
0.35
sin
β’
(
x
1
β’
(
t
k
)
)
+
(
1
+
0.15
cos
β’
(
t
k
)
)
β’
x
1
β’
(
t
k
)
+
0.3
x
2
β’
(
t
k
)
0.23
sin
β’
(
x
2
β’
(
t
k
)
)
-
0.25
x
1
β’
(
t
k
)
+
0.79
x
2
β’
(
t
k
)
]
β’
B
β‘
(
t
k
)
=
[
0.18
0.18
]
,
C
1
(
s
k
)
=
[
1.4
1
]
,
C
2
(
s
k
)
=
[
1.2
1.3
]
β’
C
3
(
s
k
)
=
[
1.3
0.9
]
,
C
4
(
s
k
)
=
[
1.3
1.3
]
wherein, x(tk)=[x1(tk) x2(tk)]T is the state vector and the fusion filter is {circumflex over (x)}CI(tk|tk)=[{circumflex over (x)}CI,1(tk|tk) {circumflex over (x)}CI,2(tk|tk)]T; sin(β‘) and cos(β‘) represent a sine function and a cosine function of ββ.
The system-related initial values are given as
x _ ( t 0 ) = [ 1.2 1.2 ] , x ^ 1 ( t 0 β t 0 ) = x ^ 2 ( t 0 β t 0 ) = x ^ 3 ( t 0 β t 0 ) = x ^ 4 ( t 0 β t 0 ) = [ 1.2 1.2 ] β’ β΅ 1 ( t 0 β t 0 ) = β΅ 2 ( t 0 β t 0 ) = β΅ 3 ( t 0 β t 0 ) = β΅ 4 ( t 0 β t 0 ) = [ 1 0 0 1 ]
The other parameters are selected as
v 1 ( s k ) = 0.6 , v 2 ( s k ) = 0.8 , v 3 ( s k ) = 0.7 , v 4 ( s k ) = 0.6 , n ~ 1 = n ~ 2 = n ~ 3 = n ~ 4 = 1 β’ Q 13 ( t k ) = Q 14 ( t k ) = 1 , Q β‘ ( t k ) = 0.6 , Ξ΅ 1 = Ξ΅ 2 = Ξ΅ 3 = Ξ΅ 4 = 0.1 , Γ² 1 = Γ² 2 = Γ² 3 = 0.5 β’ Q 21 ( t k ) = Q 24 ( t k ) = 1 , Ξ ^ 1 = Ξ ^ 2 = Ξ ^ 3 = Ξ ^ 4 = 0.16 , Ξ _ 1 = Ξ _ 2 = Ξ _ 3 = Ξ _ 4 = 0.8 β’ Q 31 ( t k ) = Q 32 ( t k ) = 1 , D 1 ( t k ) = D 2 ( t k ) = D 3 ( t k ) = D 4 ( t 4 ) = [ 0.1 0 0 0.1 ] , h = 1 β’ Q 41 ( t k ) = Q 43 ( t k ) = 1 , M 1 ( t k ) = M 2 ( t k ) β’ M 3 ( t k ) = M 4 ( t k ) = [ 0.1 0 0 0.1 ] , b = 2
log(MSEi(tk)) and log(tr{(tk|tk)}) are the logarithm of the mean square error and the logarithm of the trace of the upper bound on the local filtering error covariance of the i-th sensor node at time tk, respectively, wherein, log(β‘) represents a logarithm of ββ; MSEi(tk) represents the mean square error of the i-th sensor node at time tk. log(MSECI(tk)) and log(tr{CI(tk|tk)}) are the logarithm of the mean square error of fusion filter and the logarithm of the trace of the fusion filtering error covariance, respectively, wherein, MSECI(tk) represents the mean square error of the fusion filter at time tk. The present invention adopts the accumulated error square to demonstrate the superiority of the fusion filter, and its calculation formula is
AES
i
(
t
k
+
1
)
=
AES
i
(
t
k
)
+
ο
x
β‘
(
t
k
)
-
x
Λ
i
(
t
k
β’
β
"\[LeftBracketingBar]"
t
k
)
ο
2
AES
CI
(
t
k
+
1
)
=
AES
CI
(
t
k
)
+
ο
x
β‘
(
t
k
)
-
x
Λ
CI
(
t
k
β’
β
"\[LeftBracketingBar]"
t
k
)
ο
2
wherein, AESi(tk) represents the accumulated error square of the i-th sensor node at time tk; AESi(tk+1) represents the accumulated error square of the i-th sensor node at time tk+1; AESCI(tk) represents the accumulated error square of the fusion filter at time tk; AESCI(tk+1) represents the accumulated error square of the fusion filter at time tk+1; β₯β‘β₯ represents the norm of ββ; β₯x(tk)β{circumflex over (x)}i(tk|tk)β₯2 is the square of β₯x(tk)β{circumflex over (x)}i(tk|tk)β₯; β₯x(tk)β{circumflex over (x)}CI(tk|tk)β₯2 is the square of β₯x(tk)β{circumflex over (x)}CI(tk|tk)β₯. In particular, the average of the mean square error is introduced to evaluate further the change of fusion filtering performance under different fading probabilities, which is defined as
MSE
CI
β²
(
t
k
)
=
β
t
k
=
1
M
log
β‘
(
MSE
CI
(
t
k
)
)
/
M
,
M
=
70
is the total duration of network operation in the present invention.
Distributed Fusion Filtering Effects:
The trajectories of the actual state and its fusion filter are given in FIG. 2 and FIG. 3. The logarithm of the mean square error log(MSEi(tk)) and the logarithm of the trace of the corresponding minimum upper bound on the local filtering error covariance log(tr{(tk|tk)}) of four sensor nodes are depicted in FIG. 4. The experimental results show the effectiveness of the method proposed in the present invention.
The accumulated error square of four sensor nodes and fusion filter is given in FIG. 5. As can be seen from the figure, the accumulated error square of the fusion filter AESCI(tk) is less than the accumulated error square of the i-th sensor node AESi(tk) (i=1,2,3,4), which means that the estimation accuracy of the fusion filter is better than that of local filters. The experimental results verify the superiority of the method proposed in the present invention.
FIGS. 6-8 show log(tr{(tk|tk)}) of the sensor node i (i=1,2,3,4), log(tr{CI(tk|tk)}) and log(MSECI(tk)) of the fusion filter under different fading probabilities, respectively. It can be seen from the figures that log(tr{i(tk|tk)}), log(tr{CI(tk|tk)}) and log(MSECI(tk)) decrease with the increase of the fading probability Ξ»iu, which represents that the filtering performance of the proposed anti-eavesdropping distributed fusion filtering method will become better when the increase of Ξ»iu. Furthermore, when the fading probabilities Ξ»iu are 0.8, 0.7 and 0.3, the average mean square errors of the fusion filter are 0.4251, 3.6143 and 5.7976, i.e., when the fading probability rises from 0.3 to 0.7, the average mean square error is reduced by approximately 37%; and when it increases from 0.7 to 0.8, the reduction is approximately 88%.
In summary, the anti-eavesdropping distributed fusion filtering method over sensor network proposed in the present invention can also effectively estimate the system state in spite of simultaneously considering fading measurements and eavesdroppers.
Finally, it should be noted that the above embodiment is only the description of the technical solution of the present invention and is not limited to the scope of protection of the present invention. Although the present invention is described in detail in conjunction with the above embodiment, those having ordinary skill in the art should understand that those having ordinary skill, after reading the description of the present invention, can still make changes, modifications or equivalent replacements in the embodiment, and such changes, modifications and equivalent replacements shall fall within the scope of protection of the claims.
Claims
What is claimed is:1. An anti-eavesdropping distributed fusion filtering method for a multi-rate nonlinear system, comprising the following steps:
Step 1: establishing a dynamic model for the multi-rate nonlinear system over sensor networks:
x β‘ ( t k + 1 ) = β« ( x β‘ ( t k ) ) + B β‘ ( t k ) β’ Ο β‘ ( t k ) y i ( s k ) = Ξ i ( s k ) β’ C i ( s k ) β’ x β‘ ( s k ) + v i ( s k )
wherein, tk is a state update instant of the multi-rate nonlinear system; x(tk) is a state vector of the multi-rate nonlinear system at time tk; x(tk+1) is a state vector of the multi-rate nonlinear system at time tk+1; f(x(tk)) is a continuous and differentiable nonlinear function with a bounded second-order derivative of the multi-rate nonlinear system at time tk; B(tk) is a coefficient matrix of process noise at time tk; Ο(tk) is process noise with zero mean and covariance Q(tk) at time tk; i is a label of sensor nodes, i=1,2, . . . , N, N represents a number of sensor nodes; sk is a measurement sampling instant of sensor; x(sk) is a state vector of the multi-rate nonlinear system at time sk; yi(sk) is a measurement output signal of an i-th sensor node in the multi-rate nonlinear system at time sk; Ci(sk) is a measurement matrix of the i-th sensor node based on the multi-rate nonlinear system at time sk; vi(sk) is measurement noise of the i-th sensor node in the multi-rate nonlinear system at time sk; Ξi(sk) is used to describe a phenomenon of fading measurements;
Step 2: transforming the dynamic model for the multi-rate nonlinear system over sensor networks in Step 1 into a single-rate nonlinear system dynamic model through a prediction compensation strategy:
y Β― i ( t k ) = ( 1 - Ξ² β‘ ( t k ) ) β’ Ξ Β― i β’ C i ( t k ) β’ x Λ i ( t k β’ β "\[LeftBracketingBar]" t k - 1 ) + Ξ² β‘ ( t k ) β’ y i ( t k ) Ξ² β‘ ( t k ) = { 1 , t k = s t , t = 0 , 1 , 2 , β¦ 0 , otherwise
wherein, yi(tk) is a measurement output signal of the i-th sensor node in a single-rate nonlinear system at time tk; Ξ²(tk) is an auxiliary variable; Ci(tk) is a measurement matrix of the i-th sensor node based on the single-rate nonlinear system at time tk; {circumflex over (x)}i(tk|tkβ1) is a one-step prediction of the i-th sensor node at time tkβ1; Ξi=diag{Ξ»i1,Ξ»i2, . . . , Ξ»iny}; ny is a dimension of yi(sk); Ξ»iu is an expectation of a random variable Ξ»iu(sk), u=1,2, . . . , ny; yi(tk) is a measurement output signal of the i-th sensor node in the multi-rate nonlinear system at time tk; st is the measurement sampling instant of sensor node;
Step 3: designing an anti-eavesdropping distributed fusion filter for the single-rate nonlinear system dynamic model in Step 2; wherein
Step 3a: when a sensor node exchanges information, in order to prevent transmitted data from being eavesdropped by an eavesdropper and ensure a security of information transmission, adding artificial noise to the one-step prediction {circumflex over (x)}j(tk|tkβ1) of a sensor node j before {circumflex over (x)}j(tk|tkβ1) being sent to the sensor node i:
x Λ ij o ( t k ) = L ij ( t k ) β’ x Λ j ( t k | t k - 1 ) + ( I - L ij ( t k ) ) β’ a ij ( t k )
wherein, jβNi, Ni is a set of neighboring nodes of the i-th sensor node; {circumflex over (x)}j(tk|tkβ1) represents a one-step prediction of the j-th sensor node at time tkβ1;
x Λ ij o ( t k )
is a transmitted message from the sensor node j to the sensor node i at time tk; I is an nx-dimensional identity matrix; nx is a dimension of the state vector x(tk); aij(tk) is the artificial noise with zero mean and covariance Qij(tk) at time tk; Lij(tk) is a selection matrix at time tk;
Step 3b: when the sensor node i receives the information
x Λ ij o ( t k )
transmitted by the sensor node j, obtaining a compensated one-step prediction at time tk according to a zero-order holder compensation rule:
x Λ ij c ( t k ) = L ij ( t k ) β’ x Λ ij o ( t k ) + ( I - L ij ( t k ) ) β’ x Λ ij c ( t k - 1 )
wherein,
x Λ ij c ( t k )
is the compensated one-step prediction at time tk;
x Λ ij c ( t k - 1 )
is a compensated one-step prediction at time tkβ1;
Step 3c: designing a local distributed filter:
x Λ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) = f β‘ ( x Λ i ( t k β’ β "\[LeftBracketingBar]" t k ) ) β’ x Λ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k + 1 ) = x Λ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) + K i ( t k + 1 ) β’ ( y Β― i ( t k + 1 ) - Ξ Β― i β’ C i ( t k + 1 ) β’ x Λ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) ) + Ξ΅ i β’ β j β N i h ij ( x ^ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) - x Λ ij c ( t k + 1 ) )
wherein, {circumflex over (x)}i(tk+1|tk) represents a one-step prediction of the i-th sensor node at time tk;
x Λ ij c ( t k + 1 )
represents a compensated one-step prediction at time tk+1; {circumflex over (x)}i(tk+1|tk+1) represents a filter of the i-th sensor node at time tk+1; {circumflex over (x)}i(tk|tk) represents a filter of the i-th sensor node at time tk; f({circumflex over (x)}i(tk|tk)) represents a nonlinear function filtering form based on the single-rate nonlinear system of the i-th sensor node at time tk; Ki(tk+1) represents a local distributed filter parameter of the i-th sensor node at time tk+1; yi(tk+1) represents a measurement output signal of the i-th sensor node in the single-rate nonlinear system at time tk+1; Ci(tk+1) is a measurement matrix of the i-th sensor node based on the single-rate nonlinear system at time tk+1; Ξ΅i represents a predefined consensus parameter of the i-th sensor node; hij represents a connection coefficient between the i-th sensor node and the j-th sensor node; and
Step 3d: obtaining an anti-eavesdropping distributed fusion filter based on a local filter {circumflex over (x)}i(tk|tk) and a covariance intersection fusion criterion:
β΅ C β’ I ( t k β’ β "\[LeftBracketingBar]" t k ) = ( β i = 1 N Ο i β’ β΅ i - 1 ( t k β’ β "\[LeftBracketingBar]" t k ) ) - 1 x Λ CI ( t k β’ β "\[LeftBracketingBar]" t k ) = β΅ CI ( t k β’ β "\[LeftBracketingBar]" t k ) β’ β i = 1 N Ο i β’ β΅ i - 1 ( t k β’ β "\[LeftBracketingBar]" t k ) β’ x Λ i ( t k β’ β "\[LeftBracketingBar]" t k )
wherein, a superscript ββ1β represents an inverse of a matrix; {circumflex over (x)}CI(tk|tk) is a fusion filter at time tk; CI(tk|tk) is fusion filtering error covariance at time tk; (tk|tk) is an upper bound on a local filtering error covariance of the i-th sensor node at time tk; (tk|tk) is an inverse of a matrix
β΅ i ( t k β’ β "\[LeftBracketingBar]" t k ) ; ( β i = 1 N Ο i β’ β΅ i - 1 ( t k β’ β "\[LeftBracketingBar]" t k ) ) - 1
is an inverse of a matrix
β i = 1 N Ο i β’ β΅ i - 1 ( t k β’ β "\[LeftBracketingBar]" t k ) ;
Οi is a scalar;
Step 4: calculating an upper bound on the one-step prediction error covariance (tk+1|tk) of the i-th sensor node at time tk by solving a matrix difference equation:
β΅ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) = ( 1 + o 1 ` ) β’ A i ( t k ) β’ β΅ i ( t k β’ β "\[LeftBracketingBar]" t k ) β’ A i T ( t k ) + B β‘ ( t k ) β’ Q β‘ ( t k ) β’ B T ( t k ) + ( 1 + o ` 1 - 1 ) β’ tr β’ { M i ( t k ) β’ β΅ i ( t k β’ β "\[LeftBracketingBar]" t k ) β’ M i T ( t k ) } β’ D i ( t k ) β’ D i T ( t k )
wherein, a superscript βTβ represents the transpose of a matrix; Γ²1 is a known scaling parameter; Γ²1β1 is an inverse of Γ²1; Ai(tk) is a partial derivative of the continuous and differentiable nonlinear function f(x(tk)) corresponding to a system state at the local filter {circumflex over (x)}i(tk|tk) at time tk; Mi(tk) and Di(tk) are known error matrices obtained by Taylor series based on f(x(tk));
A i T ( t k ) , B T ( t k ) , M i T ( t k ) β’ and β’ D i T ( t k )
represent transposes of Ai(tk), B(tk), Mi(tk) and Di(tk), respectively;
Step 5: according to i(tk+1|tk) obtained in Step 4, deriving the local distributed filter parameter Ki(tk+1) of the i-th sensor node at time tk+1 by minimizing a trace of the upper bound on the local filtering error covariance:
Ki(tk+1)=Ξ΄(Ξ²(tk+1),1)(1+Γ²2)(tk+1|tk)CiT(tk+1)ΞiΞ iβ1(tk+1)
wherein,
β j ( t k + 1 ) = ( 1 + o 2 ` ) β’ Ξ Β― i β’ C i ( t k + 1 ) β’ β΅ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) β’ C i T ( t k + 1 ) β’ Ξ Β― i + tr β’ { C i ( t k + 1 ) β’ X _ i ( t k + 1 ) β’ C i T ( t k + 1 ) } β’ Ξ Λ i + R i ( t k + 1 ) β’ X _ i ( t k + 1 ) = ( 1 + o ` 3 - 1 ) β’ x Λ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) β’ x Λ i T ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) + ( 1 + o 3 ` ) β’ β΅ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k )
wherein, Ξ΄(a,b) is a Kronecker function; Ξ²(tk+1) is an auxiliary variable at time tk+1; Γ²2 and Γ²3 are known scaling parameters; Γ²3β1 is an inverse of Γ²3;
C i T ( t k + 1 )
is a transpose of Ci(tk+1);
x Λ i T ( t k + 1 β’ β "\[LeftBracketingBar]" t k )
is a transpose of {circumflex over (x)}i(tk+1|tk); Ξ iβ1(tk+1) is an inverse of Ξ i(tk+1); Ri(tk+1) is a covariance matrix of measurement noise vi(tk+1) of the i-th sensor node at time tk+1;
Step 6: by maximizing an estimation error covariance of the eavesdropper, deriving the selection matrix Lij(tk+1) at time tk+1 from the following optimization problem:
max L ij ( t k + 1 ) β L ^ ij ( t k + 1 ) t β’ r β’ { W ij ( t k + t ) }
wherein,
W ij ( t k + 1 ) = ( I - L ij ( t k + 1 ) ) β’ ( 2 β’ x Λ j ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) β’ x Λ j T ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) + Q ij ( t k - 1 ) ) β’ ( I - L ij ( t k - 1 ) )
wherein,
L Λ ij ( t k + 1 ) = { L ij 1 ( t k + 1 ) , L ij 2 ( t k + 1 ) , β¦ , L ij d i ( t k + 1 ) } , L ij 1 ( t k + 1 ) , L ij 2 ( t k + 1 ) , β¦ , L ij d i ( t k + 1 )
are diagonal matrices with elements of 0 or 1 and a sum of the diagonal elements are Γ±i; {circumflex over (x)}j(tk+1|tk) represents a one-step prediction of the j-th sensor node at time tk;
x Λ j T ( t k + 1 β’ β "\[LeftBracketingBar]" t k )
is a transpose of {circumflex over (x)}j(tk+1|tk); Qij(tk+1) is a covariance matrix of artificial noise aij(tk+1) at time tk+1;
max x f β‘ ( x )
represents that an objective function f(x) is maximized by selecting a decision variable x;
Step 7: substituting Ki(tk+1) obtained in Step 5 and Lij(tk+1) obtained in Step 6 into Step 3 to obtain the fusion filter {circumflex over (x)}CI(tk+1|tk+1) at time tk+1; determining whether tk+1 reaches a total duration M, if tk+1<M, performing Step 8, otherwise, ending;
Step 8: based on Ki(tk+1) obtained in Step 5 and Lij(tk+1) obtained in Step 6, solving for the upper bound on the local filtering error covariance (tk+1|tk+1) of the i-th sensor node at time tk+1:
(tk+1|tk+1)=(1βΞ²(tk+1))Ξi(tk+1)+Ξ²(tk+1)Ξi(tk+1)
wherein,
Ξ i ( t k + 1 ) = ( 1 + o 2 ` ) β’ β΅ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) + ( 1 + o 2 ` - 1 ) β’ Ξ΅ i 2 β’ β i β’ β j β N i h ij β’ X ij ( t k + 1 ) β’ X ij T ( t k + 1 ) Ξ i ( t k + 1 ) = ( 1 + o 2 ` ) β’ ( I - K i ( t k + 1 ) β’ Ξ Β― i β’ C i ( t k + 1 ) ) β’ β΅ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) β’ ( I - K i ( t k + 1 ) β’ Ξ Β― i β’ C i ( t k + 1 ) ) T + β¨ ( 1 + o 2 ` - 1 ) β’ Ξ΅ i 2 β’ Ο i β’ β h ij β’ X ij ( t k + 1 ) β’ X ij T ( t k + 1 ) + K i ( t k + 1 ) β’ R i ( t k + 1 ) β’ K i T ( t k + 1 ) + β¨ tr β’ { C i ( t k + 1 ) β’ X _ i ( t k + 1 ) β’ C i T ( t k + 1 ) } β’ Ξ Λ i β’ K i ( t k + 1 ) β’ K i T ( t k + 1 ) X ij ( t k + 1 ) = L ij ( t k + 1 ) β’ x Λ ij ( t k + 1 ) + ( I - L ij ( t k + 1 ) ) β’ x Λ ij c β’ ( t k + 1 ) x Λ ij c ( t k + 1 ) = x Λ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) - x Λ ij c ( t k ) , x Λ ij ( t k + 1 ) = x Λ i ( t k + 1 β’ β "\[LeftBracketingBar]" t k ) - x Λ j ( t k + 1 β’ β "\[LeftBracketingBar]" t k )
wherein, Γ²2β1 is an inverse of Γ²2;
Ξ΅ i 2
is a square of Ξ΅i; (tk+1|tk+1) is the upper bound on the local filtering error covariance of the i-th sensor node at time tkβ1;
K i T ( t k + 1 )
is a transpose of Ki(tk+1);
X ij T ( t k + 1 )
is a transpose of Xij(tk+1); (IβKi(tk+1)ΞiCi(tk+1))T is a transpose of IβKi(tk+1)ΞiCo(tk+1); ΓΈi represents a penetration of the i-th sensor node;
let tk=tk+1 and performing Step 3 until tk+1=M is satisfied; and
Step 9: estimating the multi-rate nonlinear system when simultaneously considering eavesdroppers and fading measurements for transmitted data via the sensor network, wherein when a fading probability rises from 0.3 to 0.7, an average mean square error is reduced by approximately 37%; and when the fading probability increases from 0.7 to 0.8, a reduction in the average mean square error is approximately 88%, thus improving an accuracy of a filtering performance of such problems and wherein the updated eavesdroppers and fading measurements presents a complexity of transmitted measurements of the nonlinear system to an environment monitoring computer display indicating accuracy of filtering performance.
2. The anti-eavesdropping distributed fusion filtering method for the multi-rate nonlinear system according to claim 1, wherein in Step 1, any two adjacent measurement sampling instants sk and sk+1 satisfy sk+1βsk=bh, bh is a measurement sampling period of sensor node, and b is a positive integer.
3. The anti-eavesdropping distributed fusion filtering method for the multi-rate nonlinear system according to claim 1, wherein in Step 1, Ξi(sk)=diag{Ξ»i1(sk), Ξ»i2(sk), . . . , Ξ»iny(sk)}, diag{β‘} represents a diagonal matrix composed of elements ββ, an element Ξ»iu(sk) is a random variable distributed over an interval [0,1], which meets E{Ξ»iu(sk)}=Ξ»iu and E{(Ξ»iu(sk)βΞ»iu)2}=E{({tilde over (Ξ»)}iu2(sk)}={circumflex over (Ξ»)}iu, Ξ»iu and {circumflex over (Ξ»)}iu are expectation and variance of the random variable Ξ»iu(sk), respectively, and E{β‘} is a mathematical expectation.
4. The anti-eavesdropping distributed fusion filtering method for the multi-rate nonlinear system according to claim 1, wherein in Step 3a,
L ij ( t k ) = diag β’ { β ij 1 ( t k ) , β ij 2 ( t k ) , β¦ , β ij nx ( t k ) } ,
elements
β ij 1 ( t k ) , β ij 2 ( t k ) , β¦ , β ij n x ( t k )
satisfy
β ij m β’ j β‘ ( t k ) β { 0 , 1 } β’ ( m = 1 , 2 , β¦ , n x )
and
β m = 1 n x β i β’ j m ( t k ) = n ~ i , n ~ i
is a constant that is pre-designed according to an actual demand, and βΞ£β is a summation symbol.
5. The anti-eavesdropping distributed fusion filtering method for the multi-rate nonlinear system according to claim 4, wherein, Lij(tk)β{circumflex over (L)}ij(tk),
L Λ ij ( t k ) = { L ij 1 ( t k ) , L ij 2 ( t k ) , β¦ , L i β’ j d i ( t k ) } , L ij 1 ( t k ) , L ij 2 ( t k ) , β¦ , L i β’ j d i ( t k )
are all diagonal matrices with elements of 0 or 1 and the sum of the diagonal elements is Γ±i, and
d i = C n x n i ~ , c n x n i ~
represents a number of combinations.
6. The anti-eavesdropping distributed fusion filtering method for the multi-rate nonlinear system according to claim 1, wherein in Step 3d, Οi is obtained by a following optimization problem:
min Ο i tr β’ { β΅ CI ( t k β’ β "\[LeftBracketingBar]" t k ) } s . t . β i = 1 N Ο i = 1 , Ο i β₯ 0
wherein, tr{β‘} is a trace of a corresponding matrix,
min x f β‘ ( x )
means that the objective function f(x) is minimized by selecting the decision variable x, and s.t. is an abbreviation for βsubject toβ.
7. The anti-eavesdropping distributed fusion filtering method for the multi-rate nonlinear system according to claim 1, wherein in Step 5,
Ξ΄ β‘ ( a , b ) = β’ { 1 , a = b 0 , a β b .
8. The anti-eavesdropping distributed fusion filtering method for the multi-rate nonlinear system according to claim 1, wherein in Step 8, (tk+1|tk+1) is calculated for each sensor node, such that Pi(tk+1|tk+1)β€(tk+1|t+1) holds, wherein Pi(tk+1|tk+1) is the local filtering error covariance of the i-th sensor node at time tk+1.
Images & Drawings included:
Sources:
- United States Patent and Trademark Office - verify current appl. status at the USPTOβ
Recent applications in this class:
- » 20250159017 2025-05-15
SYSTEM AND METHOD FOR DECENTRALIZED INTERNET TRAFFIC FILTERING POLICY REPORTING - » 20240430300 2024-12-26
COMMUNICATION APPARATUS, IMAGE GENERATION SYSTEM, CONTROL METHOD THEREOF, AND STORAGE MEDIUM - » 20240305665 2024-09-12
Joint time-frequency measurement in optical communication - » 20240223605 2024-07-04
Mechanism for protecting front-haul link from man-in-the-middle (MITM) attack - » 20240195835 2024-06-13
SYSTEMS AND METHODS FOR AUTOMATICALLY BLOCKING THE USE OF TRACKING TOOLS - » 20240195834 2024-06-13
MULTI-LEVEL METHODS FOR CHARACTERIZING QUANTUM COMMUNICATION CHANNELS - » 20230421606 2023-12-28
System and method for decentralized internet traffic filtering policy reporting - » 20230421605 2023-12-28
Document retention and generation at the edge - » 20230199024 2023-06-22
Systems and methods for avoiding offloading traffic flows associated with malicious data - » 20230007043 2023-01-05
Systems and methods for automatically blocking the use of tracking tools
Recent applications for this Assignee:
- » 20250038251 2025-01-30
Preparation method and application of titanium nitride fiber-enhanced quasi-solid-state electrolyte - » 18824956 2025-01-28
Secure localization method for multi-mobile robots based on network communication