Professional Documents
Culture Documents
An_Adaptive_Unscented_Kalman_Filter_Approach_for_State_Estimation_of_Nonlinear_Continuous-Discrete_System
An_Adaptive_Unscented_Kalman_Filter_Approach_for_State_Estimation_of_Nonlinear_Continuous-Discrete_System
Abstract — The paper presents algorithms for solving the One of the possible solutions to this problem is the use of
nonlinear filtering problem using an unscented Kalman filter adaptive methods for processing measurement data [7-9],
and an adaptive unscented Kalman filter. Detailed of algorithm which, along with the estimation of the state vector, can
adaptive unscented Kalman filter is provided. Step-by-step restore the statistical characteristics of noises.
schemes of filtering algorithms on the basis of which the
In this paper, an adaptive unscented Kalman Filter (AUKF)
corresponding software is developed are given. Efficiency of
nonlinear filtering algorithms application is investigated on the with noise statistic estimator is proposed for the estimation of
example of nonlinear continuous-discrete model. . Simulations nonlinear systems.
conducted on the model structure of dynamic system indicate
that the adaptive unscented Kalman filter is superior to the
conventional standard unscented Kalman filter in terms of II. PROBLEM DEFENITION
estimation accuracy and stability. Consider the following Gaussian model of stochastic
nonlinear continuous-discrete system in the state space:
Index Terms — Continuous-discrete model, nonlinear filtering,
adaptive unscented Kalman filter, noise statistic estimator. d
x(t ) = f ( x(t ), u (t )) + g (t ) w(t ), t ∈ [t0 , t N ], (1)
dt
I. INTRODUCTION y(tk +1 ) = h( x(tk +1 ),υ(tk +1 )), k = 0,1,.., N −1. (2)
need to calculate Jacobians and can achieve the second-order and is uncorrelated with w(t ) , υ ( tk +1 ) for all values of k ;
accuracy of Taylor expansion.
For the mathematical model (1), (2), taking into account
Usually the initial covariance matrices of system and
the listed a priori assumptions, it’s necessary to compare the
measurements noises are selected by analyzing some
quality of UKF and AUKF algorithms to solve the problem
empirical data or by modeling various situations. The correct of state estimation.
specification of the statistical noise parameters largely
determines the accuracy of the state vector estimation.
37
978-1-5386-7054-5/18/$31.00 ©2018 IEEE
Authorized licensed use limited to: De La Salle University. Downloaded on June 10,2024 at 12:51:22 UTC from IEEE Xplore. Restrictions apply.
XIV Международная научно-техническая конференция АПЭП – 2018
III. THEORY d
P ( t | tk ) = X S (t )WX f (t ) + X f (t )W ( X S (t ))T +
The Unscented Kalman Filter belongs to a bigger class of dt
filters called Sigma-Point Kalman Filters, which are using + g (t )Qˆ (t ) gT (t )
the statistical linearization technique. This technique is used
to linearize a nonlinear function of a random variable with initial conditions xˆ(tk | tk ), P (tk | tk ) and Qˆ (t k ) for
through a linear regression between points drawn from the tk ≤ t ≤ tk +1 .
prior distribution of the random variable.
Update:
The following AUKF uses the Sage-Husa algorithm to
estimate the unknown noise [7], and estimates and corrects • Find sigma points X S (tk +1 )
the process noise statistical characteristics in real time to xˆ (tk +1 | tk ) 0
reduce the state estimation error. xˆ(t | t ) D
Initialization: k +1 k 1
• Set initial values
α = 0.0001, β = 2, b = 0.998,
X S (tk +1 ) = xˆ (tk +1 | tk ) + ( L + λ ) DL ,
xˆ(t0 | t0 ) = x (t0 ), P(t0 | t0 ) = P (t0 ), Q(t0 ), R (t1 ) . xˆ (tk +1 | tk ) −D
• Calculate 1
λ
λ = α 2 L − L; W0( m ) = , xˆ(tk +1 | tk ) − DL
L+λ
where Di – i -th row of the lower triangular matrix obtained
λ
W0( c ) = + (1 − α 2 + β ), by the Cholesky decomposition P (tk +1 | tk ) .
L+λ
1 • Calculate
Wi ( m ) = Wi ( c ) = , i = 1,.., 2 L.
2( L + λ ) Y h (tk +1 ) = h( X 0s (tk +1 ),0), , h( X 2sL (tk +1 ),0) .
ωm = [W0( m ) ,...,W2(Lm ) ], • Find
1− b
W = (I − [ωm ,...,ωm ]) × diag (W0(c) ,...,W2(Lc) ) × (I − [ωm ,...,ωm ])T , dk = ;
1 − b k +1
where L – dimension state vector.
Time update k = 0, N − 1 ε (tk +1 ) = y(tk +1) − Y h (tk +1)ωmT ;
Predict: Rˆ (tk +1 ) = (1 − d k ) Rˆ (tk ) + d k (ε (tk +1 )ε T (tk +1 ) −
• Calculate sigma points X S (t ) in accordance with 2L
38
Authorized licensed use limited to: De La Salle University. Downloaded on June 10,2024 at 12:51:22 UTC from IEEE Xplore. Restrictions apply.
2018 14th International Scientific-Technical Conference APEIE – 44894
1δk ,i ,t j ∈{25,29,45,46,52,53,64,88,95,96} ;
- R ( tk +1 ) =
0.02δk ,i ,otherwise,
(test 2),
1δk ,i ,t j ∈{31,32,33,34,35,36,37,38,39,40} ;
- R ( tk +1 ) =
y ср ( t k + 1 ) , yср ( tk +1 )
*
0.02δk ,i ,otherwise, Fig. 2. Graphical representation and
y ср ( t k + 1 ) , yср ( tk +1 )
*
TABLE I Fig. 3. Graphical representation and
NUMERICAL RESULTS OF NONLINEAR FILTERING
ŷср ( tk +1 ) (test 2, UKF).
UKF AUKF
39
Authorized licensed use limited to: De La Salle University. Downloaded on June 10,2024 at 12:51:22 UTC from IEEE Xplore. Restrictions apply.
XIV Международная научно-техническая конференция АПЭП – 2018
ACKNOWLEDGMENT
The work is executed under the auspices of the Ministry of
education and science of the Russian Federation (project №
№ 2.7996.2017/8.9).
V. CONCLUSION
In this paper, an AUKF with noise statistic estimator is
proposed for the estimation of nonlinear continuous-discrete
systems. This improves the robustness of conventional UKF
with respect to the variable noise distribution in the real
system. Simulations between the conventional UKF and the
AUKF are carried out. The results prove that, in the situation
y ср ( t k + 1 ) , yср ( tk +1 ) of uncertain or time-varying noise statistic, the adaptive UKF
*
Fig. 4. Graphical representation and
outperforms the conventional UKF in terms of the fast
ŷср ( tk +1 ) (test 2, AUKF). convergence and state estimation accuracy by applying noise
statistic estimator to tune the noise statistic.
Both theoretical analysis and experimental results show
that the proposed AUKF algorithm is superior to the standard
UKF algorithm in estimating performance and stability.
REFERENCES
[1] M.A. Ogarkov. Methods for statistical estimate of the parameters of
random processes. Rev. and ext. Moscow: Energoatomizdat, 1980, 208
p.
[2] Eric A. Wan, Rudolph van der Merwe. The Unscented Kalman Filter
for Nonlinear Estimation. 2000 IEEE Conference: Adaptive Systems for
Signal Processing, Communications, and Control Symposium 2000.
AS-SPCC – February, 2000. P. 6.
[3] S. Sarkka. On Unscented Kalman Filtering for State Estimation of
Continuous-Time Nonlinear Systems. IEEE Transactions on Automatic
y ср ( t k + 1 ) , yср ( tk +1 )
* Control – October, 2007. P. 11.
Fig. 5. Graphical representation and [4] Ishihara S, Yamakita M. Continuous-discrete robust UKF for nonlinear
systems with parameter uncertainties // 48th ISCIE International
ŷср ( tk +1 ) (test 3, UKF). Symposium on Stochastic Systems Theory and Its Applications.
Fukuoka – November, 2016. 6 p.
[5] M. Verhaegen, V. Verdult. Filtering and system identification: a least
squares approach: Cambridge University Press, 2007, 422 p.
[6] S.J.Julier and J.K.Uhlmann. A New Extension of the Kalman Filter to
Nonlinear Systems. Proc. Of AeroSense: The 11th Int. Symp. on
Aerospace / Defence Sensing, Simulation and Controls.,1997.
[7] H. Wang, G. Fu, J. Li. An Adaptive UKF Based SLAM Method for
Unmanned Underwater Vehicle. Hindawi Publishing Corporation
Mathematical Problems in Engineering, 2013, Vol.13, 13 p.
[8] A. Mohamed, K. Schwarz. Adaptive Kalman filtering for INS/GPS.
Journal of Geodesy, 1999, Vol.73(4), pp. 193-203.
[9] V.Chubich, O. Chernikova, A. Dolgov. Robust estimation of parameters
of gaussian models linear discrete systems based on hybrid filter.
Bulletin of Irkutsk state University, 2017, Vol.21, N.4, pp.100-107. (In
Russian)
40
Authorized licensed use limited to: De La Salle University. Downloaded on June 10,2024 at 12:51:22 UTC from IEEE Xplore. Restrictions apply.