Sampling error distribution for the ensemble Kalman filter update step
Multiple Adaptive Fading Schmidt-Kalman Filter for Unknown Bias
Transcript of Multiple Adaptive Fading Schmidt-Kalman Filter for Unknown Bias
Research ArticleMultiple Adaptive Fading Schmidt-Kalman Filterfor Unknown Bias
Tai-Shan Lou, Zhi-Hua Wang, Meng-Li Xiao, and Hui-Min Fu
School of Aeronautical Science and Engineering, BeiHang University, Beijing 100191, China
Correspondence should be addressed to Zhi-Hua Wang; [email protected]
Received 24 September 2014; Accepted 12 November 2014; Published 24 November 2014
Academic Editor: Zheng-Guang Wu
Copyright Β© 2014 Tai-Shan Lou et al. This is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Unknown biases in dynamic and measurement models of the dynamic systems can bring greatly negative effects to the stateestimates when using a conventional Kalman filter algorithm. Schmidt introduces the βconsiderβ analysis to account for errorsin both the dynamic and measurement models due to the unknown biases. Although the Schmidt-Kalman filter βconsidersβ thebiases, the uncertain initial values and incorrect covariance matrices of the unknown biases still are not considered. To solve thisproblem, a multiple adaptive fading Schmidt-Kalman filter (MAFSKF) is designed by using the proposed multiple adaptive fadingKalman filter to mitigate the negative effects of the unknown biases in dynamic or measurement model. The performance of theMAFSKF algorithm is verified by simulation.
1. Introduction
An underlying assumption of the Kalman filter is that thedynamic andmeasurement equations can be accuratelymod-eled without any colored noise or unknown biases. However,in practice, these dynamic and measurement models includesome additional biases, which always bring greatly negativeeffects to the state estimate.
There are many methodologies to deal with theseunknown biases. Ignoring them and augmenting them toestimate are two common approaches. Based on the sen-sitivity to the unknown bias, some techniques have beenproposed, such as Ξ
βfiltering [1, 2], set-valued estimation
[3], and Schmidt-Kalman filter (SKF) [4]. Schmidt proposeda βconsiderβ analysis, which is the cornerstone of the SKF,to account for errors in both the dynamic and measurementmodels due to the unknown biases when the biases areconsidered as constants and remain unchanged [4]. Based ona minimum variance approach, the key idea of the SKF isthe βconsiderβ analysis that the preestimated bias covarianceis formulated to update the state and covariance estimates,but these biases themselves are not estimated directly. Theβconsiderβ approach is especially useful when the unknown
biases are low observable or when the extra computationalpower to estimate them is not worth [5].
After Schmidt, the βconsiderβ approach for parametershas received much attention in recent years. The SKF is alsocalled the consider Kalman filter (CKF) after its developer.Jazwinski provides the detailed derivation of the CKF inhis book [6]. Subsequently, Tapley et al. amply descript theCKF and derivate a different formulation [7]. Zanetti andSouza introduce the UDU formulation into the SKF andprovide a numerically stability, recursive implementation ofthe UDU SKF [8]. Bierman analyzes the effects on filteringaccuracy of the unestimated biases and incorrect a prioricovariance statistics and proposes a sensitivity matrix toevaluate them [9]. Woodbury et al. give novelty insight intoconsidering biases in the measurement model and verifythe negative effect of the errors in the initial parameter andcovariance estimates [5, 10]. Chee and Forbes propose anorm-constrained consider Kalman filtering by taking intoaccount the constraint on the state estimate and apply it to anonlinear attitude estimation problem [11].
However, how to mitigate these negative effects from theinitial state and covariance values of the unknown biasesin the SKF has not attracted much attention. In fact, when
Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2014, Article ID 623930, 8 pageshttp://dx.doi.org/10.1155/2014/623930
2 Mathematical Problems in Engineering
the inaccurate initial and covariance values, that is to say thebiases are not accurately modeled, are used to update thestate and covariance estimates, the accuracy of the state andcovariance estimates may greatly degrade. Fortunately, theadaptive technique is proposed to improve the convergenceof the filtering. As a member of the adaptive Kalman filteringalgorithms, the adaptive fading Kalman filtering algorithmis proposed to use a single adaptive fading factor (FF) as amultiplier to the dynamic or measurement noise covariancewhen the information about the dynamic or measurementmodel is incomplete [12β15]. Then, to consider the complexsystems with multivariable, a single fading factor is notsufficiently used, and so the multiple fading factor, which isthe footstone of multiple adaptive fading Kalman filtering(MAFKF), is proposed to reflect corrective effects of themultivariable in filtering [16β19]. But in the MAFKF themultiple fading factors are only used as a multiplier for thelast posteriori covariance of the states, and the method for thewhole priori covariance of the states is not considered untilnow.
To consider the incomplete information from both thecovariance of the states and noises, the MAFKF is proposedto use themultiple fading factor as a multiplier on the outsideof the whole priori error covariance equation. The proposedMAFKF not only considers the uncertainty of the models butalso adjusts the covariance of inaccurate modeled noises. Inaddition, the multiple fading factors are derived by one-stepapproximate algorithm to decrease the computational com-plexity in the MAFKF algorithm.Then, the multiple adaptivefading Schmidt-Kalman filter (MAFSKF) is designed byusing the aboveMAFKF tomitigate the negative effects of theuncertain parameters in dynamic or measurement model.
This paper is organized as follows. First, the problemstatement with the unknown biases is given. Second, theMAFKF algorithm is proposed to compensate the effect ofinaccuracy information covariance. Third, the MAFSKF isdesigned to mitigate the negative effects of the unknownbiases. Finally, the performance of the MAFSKF algorithm isverified by simulation and the results are discussed as well.
2. Problem Statement
Consider a linear discrete dynamic systemwith the unknownbiases as follows:
xπ+1
= Ξ¦π+1|π
xπ+Ξ¨π+1|π
pπ+ Gπwπ
(1a)
zπ= Hπxπ+ Nπbπ+ kπ, (1b)
where xπis the π Γ 1 state vector and z
πis the π Γ 1
measurement vector. Ξ¦π+1|π
and Ξ¨π+1|π
are the state andbias transition matrices, G
πis the coefficient matrix of the
process noise, Hπis the measurement matrix, and N
πis
the measurement bias transition matrix. pπis referred to
as the ππΓ 1 dynamical bias vector and b
πis called the
ππΓ 1 measurement bias vector. w
πand kπare independent
zero-meanGaussian noise processes and their covariance are,respectively,Q
πand R
π. They satisfy
πΈ [wπwππ] = Q
ππΏππ,
πΈ [kπkππ] = RππΏππ,
πΈ [wπkππ] = 0,
(2)
where πΏππis the Kronecker delta function andQ
π> 0, R
π> 0.
Here, the biases pπand b
π, which are considered as
unknown constants and remain the same in filtering, aremodeled as
pπ+1
= pπ, p0= p,
bπ+1
= bπ, b0= b.
(3)
The initial states x0and biases p
0and b
0are assumed to
be independent of the Gaussian noise {wπ} and {k
π} and be
Gaussian random variables with
πΈ [x0] = x0, πΈ [(x
0β x0) (x0β x0)π
] = P0> 0,
πΈ [p0] = p0, πΈ [(p
0β p0) (p0β p0)π
] = Qπ0> 0,
πΈ [b0] = b0, πΈ [(b
0β b0) (b0β b0)π
] = Qπ0> 0,
πΈ [(x0β x0) (p0β p0)π
] = Cπ₯π0,
πΈ [(x0β x0) (b0β b0)π
] = Cπ₯π0.
(4)
Based on the assumption that the stochastic informationof the unknown biases is incomplete, a multiple adaptivefading Schmidt-Kalman filter is designed to overcome theproblem with the unknown biases.
3. MAFKF Algorithm
Consider the linear discrete stochastic system as follows:
xπ+1
= Ξ¦π+1|π
xπ+ Gπwπ, (5a)
zπ= Hπxπ+ kπ. (5b)
If the system is observable, the optimal estimate is givenby the conventional Kalman filter [7]. Unfortunately, theinformation is always incomplete in practice, and this leadsthe filter to βlearn the wrong state too wellβ [20]. To compen-sate the effects of the incomplete information, the adaptivefading Kalman filter (AFKF) is proposed to overcome theproblem [12, 13]. When the older data from the currentestimate are no longer meaningful due to the erroneousmodel, the negative effects of these data are mitigated by theAFKF. Between theAFKF and the conventional Kalman filter,the big difference is that a constant fading factor is insertedinto the a priori error covariance equation. There are three
Mathematical Problems in Engineering 3
representative types with a single fading factor to be assigned[12, 13, 15, 21],
Pπ+1|π
= ππΞ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ, (6a)
Pπ+1|π
= Ξ¦π+1|π
PπΞ¦π
π+1|π+ ππGπQπGππ, (6b)
Pπ+1|π
= ππ(Ξ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ) . (6c)
However, only one constant fading factor cannot βweightβthe covariance of all states, and the optimal filtering cannotbe guaranteed, especially for the complicated multivariablesystems. To overcome the shortcomings of the single fadingfactor, Zhou et al. [16] proposed a suboptimal multiple fadingextendedKalmanfilter by using the error covariance equationof (6a), in which the single fading factor π
πis substituted by
a multiple fading factor matrix Sπ, and the orthogonality of
the residual errors is remained. Zhou et al. also gave one-step approximation algorithm of the multiple fading factorand verified the affectivity of the multiple fading factor infiltering. The equation of (6b) and (6c) was considered fora single fading factor in the literature [13, 16]. But for themultiple fading factor, no researchers consider the last twoequations. To consider the incomplete information fromboththe covariance of the states and noises, the multiple fadingfactor should be inserted on the outside of the a priori errorcovariance equation. Hence, based on (6c) as the comment inthe literature [17], the proposed MAFKF is defined as
Pπ+1|π
= Sπ+1(Ξ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ) , (7)
where Sπ+1
= diag{π1,π+1
, π2,π+1
, . . . , ππ,π+1
}; ππ,π+1
β₯ 1 (π =
1, 2, . . . , π) is the multiple fading factor.In the conventional Kalman filter, the predicted residual
vector can be expressed asπΎπ+1
= zπ+1βHπ+1
xπ+1|π
(8)
and the corresponding innovation covariance matrix can becalculated as
Ξ©π+1
= πΈ [πΎπ+1πΎπ
π+1] = H
π+1Pπ+1|π
Hππ+1+ Rπ+1, (9)
where Pπ+1|π
is a priori error covariance of the linear Kalmanfilter.
In the optimal linear Kalman filter, there is an orthogonalprinciple that the predicted residual sequence {πΎ
π} ismutually
orthogonal when the optimal gain matrix is calculated online[16].The optimal gainmatrix is obtained in the linear Kalmanfilter by minimizing the following equation:
πΈ [(xπ+1β xπ+1) (xπ+1β xπ+1)π
] , π = 0, 1, 2, . . . , (10)
and then the following equation is satisfied:
πΈ [πΎπ+π+1πΎπ
π+1] = 0, π = 0, 1, 2, . . . , π = 1, 2, 3, . . . . (11)
Substituting (8) into the left formula of (11), the result can beobtained as follows:πΈ [πΎπ+π+1πΎπ
π+1]
= Hπ+π+1Ξ¦π+π+1|π+π
[I β Kπ+π
Hπ+π] β β β Ξ¦
π+3|π+2
Γ [I β Kπ+2
Hπ+2]Ξ¦π+2|π+1
Ξπ+1, π = 1, 2, 3, . . . ,
(12)
where Ξπ+1
is defined as (for all π = 0, 1, 2, 3, . . .)
Ξπ+1
= Pπ+1|π
Hππ+1β Kπ+1Ξ©π+1. (13)
Substituting the optimal gain matrix Kπ+1
=
Pπ+1|π
Hππ+1[Hπ+1
Pπ+1|π
Hππ+1
+ Rπ+1]β1 of the linear Kalman
filter into (13), Ξπ+1
is identically zero, and this means that(12) is identical to zero, too. The orthogonal principle is rightwhen the optimal gain matrix is inserted.
In practice, the dynamic model of the stochastic systemalways is partially known, and so the real covariance matrixmay be increased by the unknown information, and it isdifferent from the theoretical covariance Ξ©
π+1in (9). Thus,
the real autocovariance matrix πΈ[πΎπ+π+1πΎπ
π+1] may not be
identically zero. For (12), if the multiple fading factor in (7)is chosen so that Ξ
π+1= 0, then the gain matrix K
π+1is
optimal. From the above, it can be seen that ifKπ+1
is optimal,Ξπ+1
= 0 in (12), and if Ξπ+1
= 0, Kπ+1
is optimal. The basicidea to design the adaptive fading filtering is obtained fromaforementioned analysis.
Hence, the optimality of the Kalman filter can be evalu-ated by the following function constructed:
π (ππ+1) =
π
β
π=1
π
β
π=1
Ξ2
ππ,π+1, (14)
where Ξπ+1
= (Ξππ,π+1
)ππ
and ππ+1
= [π1,π+1
, π2,π+1
, . . . ,
ππ,π+1
]π. π(π
π+1) describes the distance to the optimal esti-
mate in linear Kalman filter. When π(ππ+1) is minimum, a
suboptimal estimate, which is the most close to the optimalestimate, will be obtained. Hence, we can obtain the multiplefading factor S
πby minimizing (14) as follows:
minππ
π (ππ) . (15)
Obviously, (15) can be solved by using any unconstrainedmultivariate nonlinear programming methods. However,finding the optimal solution is not suitable for the onlinestate estimate [16]. Hence, a one-step approximate algorithmis proposed to obtain the multiple fading factors S
πfor the
online calculating.Here, when the a priori characters of the system are
roughly known, we can assume that
π1,π+1
: π2,π+1
: β β β : ππ,π+1
= π½1: π½2: β β β : π½
π(16)
and then set
ππ,π+1
= π½πππ+1, π = 1, 2, . . . , π, (17)
where π½πβ₯ 1 is the constant from the prognosis to the state
and ππ+1
is the undetermined factor.Substituting the Kalman gain matrix K
π+1=
Pπ+1|π
Hππ+1[Hπ+1
Pπ+1|π
Hππ+1
+ Rπ+1]β1 into the Ξ
π+1= 0
generates
Pπ+1|π
Hππ+1{I β [H
π+1Pπ+1|π
Hππ+1+ Rπ+1]β1
Ξ©π+1} = 0.
(18)
It is obvious that one sufficient condition to establish (18) is
[Hπ+1
Pπ+1|π
Hππ+1+ Rπ+1]β1
Ξ©π+1
= I (19)
4 Mathematical Problems in Engineering
or
Hπ+1
Pπ+1|π
Hππ+1
= Ξ©π+1β Rπ+1. (20)
Substituting (7) into (20) and reorganizing it gives
Hπ+1
Sπ+1(Ξ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ)Hππ+1
= Ξ©π+1β Rπ+1.
(21)
From the right part of (21), it is seen that the multiple fadingfactor S
π+1is valid when Ξ©
π+1β Rπ+1
> 0 is satisfied[16]. For the measurement covariance matrix R
π+1> 0, a
softening factor π β₯ 1, which is usually given by experience,is introduced to weaken the excessive adjust of the multiplefading factor and smooth the state estimates. Hence, (21) isrestructured as
Hπ+1
Sπ+1(Ξ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ)Hππ+1
= Ξ©π+1β πRπ+1.
(22)
Based on the property of commutative matrices in traceoperator tr[π΄π΅] = tr[π΅π΄], the trace of both sides in (22) iscalculated as
tr [Sπ+1(Ξ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ)Hππ+1
Hπ+1] = tr [Ξ©
π+1β πRπ+1] .
(23)
Simplify (23) as
tr [Sπ+1
Mπ+1] = tr [O
π+1] , (24)
where
Mπ+1
= (Ξ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ)Hππ+1
Hπ+1, (25)
Oπ+1
= Ξ©π+1β πRπ+1. (26)
In fact, the real residual covariance matrixΞ©π+1
in (26) isunknown but can be evaluated by the following equation [16]:
Ξ©π+1
=
{{
{{
{
πΎ1πΎπ
1, π = 1
πΞ©π+ πΎπ+1πΎπ
π+1
1 + π, π > 1,
(27)
where π is a forgetting factor, which is 0 < π β€ 1.Substituting (17) into (24) gives
tr{{{
{{{
{
[[[
[
π½1ππ+1
π½2ππ+1
dπ½πππ+1
]]]
]
Mπ+1
}}}
}}}
}
= tr [Oπ+1]
(28)
and then ππ+1
is calculated as
ππ+1
=tr [Oπ+1]
βπ
π=1π½πMππ,π+1
. (29)
Synthesizing condition ππ,π+1
β₯ 1 and (17), (29) gives
ππ,π+1
= max {1, π½πππ+1} , π = 1, 2, . . . , π. (30)
Algorithm 1 (one-step approximate MAFKF). A discrete-timemultiple adaptive fadingKalmanfilter is proposed by thefollowing equations when the information about the linearstochastic system is incomplete:
xπ+1|π
= Ξ¦π+1|π
xπ, (31a)
Pπ+1|π
= Sπ+1(Ξ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ) , (31b)
Kπ+1
= Pπ+1|π
Hππ+1[Hπ+1
Pπ+1|π
Hππ+1+ Rπ+1]β1
, (31c)
xπ|π= xπ|πβ1
+ KππΎπ+1, (31d)
Pπ= (I β K
πHπ)Pπ|πβ1
, (31e)
where
πΎπ+1
= zπ+1βHπ+1
xπ+1|π
,
Sπ+1
= diag {π1,π+1
, π2,π+1
, . . . , ππ,π+1
} ,
ππ,π+1
= max {1, π½πππ+1} , π = 1, 2, . . . , π,
ππ+1
=tr [Oπ+1]
βπ
π=1π½πMππ,π+1
,
Mπ+1
= (Ξ¦π+1|π
PπΞ¦π
π+1|π+ GπQπGππ)Hππ+1
Hπ+1,
Oπ+1
= Ξ©π+1β πRπ+1,
Ξ©π+1
=
{{
{{
{
πΎ1πΎπ
1, π = 1
πΞ©π+ πΎπ+1πΎπ
π+1
1 + π, π > 1.
(32)
To consider the difference between the states and themultisource incomplete information, the MAFKF algorithmis proposed to use the multiple fading factor as a multiplierfor the whole a priori covariance P
π+1|πof the states and
mitigate the negative effects of the uncertainties. Comparedto the single adaptive fading Kalman filter, the MAFKF isintroduced into the multiple fading factor and adjusts eachcomponent of the state vector by different fading factor toperform better. In addition, the multiple fading factor isderived by one-step approximate algorithm to decrease thecomputational complexity.
The asymptotical stability of the proposed MAFKF iseasily proved in the literature [15], by using the results in theliterature [22β24].
Remarks. The proportionality factor π½πof the multiple fading
factor Sπcan be designed with the a priori knowledge of the
states before the filtering [16].
4. MAFSKF for the Unknown Biases
The unknown biases in the problem statement (1a) and (1b)have a greatly negative impact on the filter accuracy andeven result in filter divergence [6]. In the SKF algorithm, thecovariance of unknown biases is used to update the state andcovariance estimates but is not estimated directly.However, as
Mathematical Problems in Engineering 5
the most important part in the SKF, the unknown covariancematrices Qπ
0and Qπ
0and the uncertain initial values of the
unknown biases are still not considered. To consider thenegative effects in the SKF filtering from this incompleteinformation, the MAFKF is proposed to solve this problem.Based on the two aforementioned aspects, the MAFSKF isdesigned by using the MAFKF and the conventional SKF.
Similarly to (8) and (9) in the conventional Kalmanfilter, the predicted residual vector and the correspondinginnovation covariance matrix in SKF can be expressed as
πΎπ+1
= zπ+1βHπ+1
xπ+1|π
β Nπ+1
b, (33)
Ξ©π+1
= πΈ [πΎπ+1πΎπ
π+1] = H
π+1Pπ+1|π
Hππ+1+ Nπ+1
Cπππ+1|π
Hππ+1+Hπ+1
Cππ+1|π
Nππ+1
+ Nπ+1
Qπ0Nππ+1+ Rπ+1,
(34)
where the unknown real residual covariance matrix Ξ©π+1
iscalculated by (27).
The incomplete information, coming from the unknownbiases in models, can be obtained from (27), (33), and (34).So the multiple fading factor can be calculated from (27) and(34) and used to compensate the corresponding covariancematrix and autocovariance matrix.
The SKF algorithm is derived as follows [4, 25]. First, theunknown biases are augmented into the states and the newaugmented system is produced. Second, the standard linearKalman filter is derived from the augmented system. Third,the estimation equations for the biases are thrown away, butthe covariance between the states and biases is retrained.From the recursions in SKF [25], the a priori error covarianceincludesPπ+1|π
= Ξ¦π+1|π
PπΞ¦π
π+1|π+Ξ¦π+1|π
CππΞ¨π
π+1|π+Ξ¨π+1|π
CπππΞ¦π
π+1|π
+Ξ¨π+1|π
Qπ0Ξ¨π
π+1|π+ GπQπGππ,
Cππ+1|π
= Ξ¦π+1|π
Cππ+Ξ¨π+1|π
Qπ0,
Cππ+1|π
= Ξ¦π+1|π
Cππ.
(35)
From the recursive process, the three above covari-ance matrices should be adjusted by the multiple fad-ing factor S
π+1, for the incomplete information of the
unknown biases. The multiple fading factor Sπ+1
=
diag{π1,π+1
, π2,π+1
, . . . , ππ,π+1
} is calculated for the augmentedsystem like the proposed MAFKF. Under assumption thatππ,π+1
= max{1, π½πππ+1}, π = 1, 2, . . . , π, π
π+1is calculated as
in (29), and Oπ+1
still remains the same, but Mπ+1
in (29) ischanged into
Mπ+1
= (Ξ¦π+1|π
PπΞ¦π
π+1|π+Ξ¦π+1|π
CππΞ¨π
π+1|π
+Ξ¨π+1|π
CπππΞ¦π
π+1|π
+ Ξ¨π+1|π
Qπ0Ξ¨π
π+1|π+ GπQπGππ)Hππ+1
Hπ+1
+Ξ¦π+1|π
CππNππ+1
Hπ+1.
(36)
From the analysis above, a multiple adaptive fadingSchmidt-Kalman filter is proposed to mitigate the negativeeffects of the unknown biases.
Algorithm 2 (one-step approximate MAFSKF). Based on theone-step approximateMAFKF algorithm, amultiple adaptivefading Schmidt-Kalman filter is proposed by the followingequations when the information about the linear stochasticsystem is incomplete:
xπ+1|π
= Ξ¦π+1|π
xπ+Ξ¨π+1|π
p, (37a)
Pπ+1|π
= Sπ+1(Ξ¦π+1|π
PπΞ¦π
π+1|π+Ξ¦π+1|π
CππΞ¨π
π+1|π
+Ξ¨π+1|π
CπππΞ¦π
π+1|π
+ Ξ¨π+1|π
Qπ0Ξ¨π
π+1|π+ GπQπGππ) ,
Cππ+1|π
= Sπ+1(Ξ¦π+1|π
Cππ+Ξ¨π+1|π
Qπ0) ,
Cππ+1|π
= Sπ+1(Ξ¦π+1|π
Cππ) ,
(37b)
Kπ+1
= [Pπ+1|π
Hππ+1+ Cππ+1|π
Nππ+1]Ξβ1
π+1, (37c)
Ξπ+1
= Hπ+1
Pπ+1|π
Hππ+1+ Nπ+1
Cπππ+1|π
Hππ+1
+Hπ+1
Cππ+1|π
Nππ+1
+ Nπ+1
Qπ0Nππ+1+ Rπ+1,
(37d)
xπ+1
= xπ+1|π
+ Kπ+1πΎπ+1, (37e)
Pπ+1
= (I β Kπ+1
Hπ+1)Pπ+1|π
β Kπ+1
Nπ+1
Cπππ+1|π
,
Cππ+1
= (I β Kπ+1
Hπ+1)Cππ+1|π
,
Cππ+1
= (I β Kπ+1
Hπ+1)Cππ+1|π
β Kπ+1
Nπ+1
Qπ0,
(37f)
whereπΎπ+1
= zπ+1βHπ+1
xπ+1|π
β Nπ+1
b,
Sπ+1
= diag {π1,π+1
, π2,π+1
, . . . , ππ,π+1
} ,
ππ,π+1
= max {1, π½πππ+1} , π = 1, 2, . . . , π,
ππ+1
=tr [Oπ+1]
βπ
π=1π½πMππ,π+1
,
Oπ+1
= Ξ©π+1β πRπ+1,
Ξ©π+1
=
{{
{{
{
πΎ1πΎπ
1, π = 1
πΞ©π+ πΎπ+1πΎπ
π+1
1 + π, π > 1.
(38)
5. Simulation Results and Analysis
To evaluate the performance of the proposed MAFSKFalgorithm, the spacecraft attitude tracing system with
6 Mathematical Problems in Engineering
0 20 40 60 80 100 120 140 160 180 2000
0.1
0.2
0.3
0.4
0.6
0.5
Step
p
Unk
now
n bi
asp
Figure 1: Values of the unknown bias.
the gyroscope as a measurement sensor is considered [26].The spacecraft attitude tracking system, which is mainlyused to enhance and track the spacecraft drift signal, iscorrupted by the unknown bias. The corresponding discrete-time dynamic stochastic system is expressed as
xπ+1
= [0 1
β0.85 1.70] xπ+ [
0.0129
β1.2504] pπ+ [0
1]wπ, (39)
zπ= [0 1] x
π+ kπ. (40)
As in the literature [21], we assumed that x = [π₯1, π₯2]π,wπβΌ π(0, 0.01
2
), and kπβΌ π(0, 0.01
2
). The true state x0=
[2, 1]π and the unknown bias p
0βΌ π(0, 0.5
2
) were added tothe dynamic equation (39). The initial estimates of the statex0= [1.8, 0.9]
π, and bias p0= 0. In simulation, the dynamic
system is disturbed by some external disturbances, and thevalues of the unknown biases p
πare plotted in Figure 1. The
MAFSKF and SKF use Qπ = 0.001 Γ 0.52 as the covarianceof the bias p, which is one thousandth of the real covarianceQπ = 0.5
2. The constant π½ in MAFSKF algorithm is set asπ½1= 1 and π½
2= 100, the softening factor π is set as π = 1, and
the forgetting factor is π = 0.95.Each single run lasts for 200 samples and 100Monte Carlo
runs are performed.The rootmean squared errors (RMSE) ofthe state estimate are calculated to compare the performanceof the MAFSKF algorithm and the SKF algorithm at eachepoch. The simulation results are shown in Figures 2β5.
Figure 2 shows the time evolution of the multiple fadingfactor in simulation. From Figure 2, we can see that thefading factors of the states become larger in order to recoverthe filter from divergence, and the second fading factor isgreatly larger than the first one because the second statechanges largely. Figures 3 and 4 show the RMSEs of thestate estimates both π₯1 and π₯2 by using the conventionalSKF and MAFSKF algorithm, respectively. The RMSEs ofthe MAFSKF algorithm on its own are given in Figure 5 towell show the results. From Figures 3 and 4, it is obviouslyseen that the MAFSKF algorithm can adapt these unknownbiases and well track the true states, compared to the SKF. As
0 20 40 60 80 100 120 140 160 180 2001
1.52
2.53
3.54
Fadi
ng fa
ctor
FF1
0 20 40 60 80 100 120 140 160 180 2000
50100150200250300350400
Step
Step
Fadi
ng fa
ctor
FF2
Figure 2: Time evolution of the multiple fading factor.
0 20 40 60 80 100 120 140 160 180 2000
0.01
0.02
0.03
0.04
0.05
0.06
Step
SKFMAFSKF
x1
RMSE
Figure 3: RMSE of π₯1 for 100 random simulation runs.
a result, the performance of the MAFSKF algorithm is betterthan the SKF when the information of the unknown biases isincomplete.
6. Conclusions
In this paper, the multiple adaptive fading Schmidt-Kalmanfilter is presented to mitigate the negative effects of theunknown biases in dynamic or measurement model. Inpractice situations, the dynamic and measurement modelsinclude some additional unknown biases, which always bringgreatly negative effects to the state estimates. Although theSchmidt-Kalman filter βconsidersβ the biases, the uncer-tain initial values and incorrect covariance matrices of theunknownbiases still are not considered. To solve the problem,
Mathematical Problems in Engineering 7
0 20 40 60 80 100 120 140 160 180 2000
0.02
0.04
0.06
0.08
0.1
0.12
Step
SKFMAFSKF
x2
RMSE
Figure 4: RMSE of π₯2 for 100 random simulation runs.
0 20 40 60 80 100 120 140 160 180 2000
0.0050.01
0.0150.02
0.0250.03
MAFSKF
MAFSKF
0 20 40 60 80 100 120 140 160 180 200789
101112
Step
Step
x1
RMSE
x2
RMSE
Γ10β3
Figure 5: RMSE for MAFSKF.
the MAFKF is proposed to adjust the covariance of the statesand noise by using the multiple fading factors as a multiplieron the outside of the a priori covariance equation when theinformation about the dynamic or measurement model isincomplete. Then, the MAFSKF is designed based on theMAFKF. Numerical simulation shows that the MAFSKF canmitigate the negative effects of incorrect covariance matricesof the unknown biases compared to the SKF.
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper.
Acknowledgments
The work described in this paper was supported by theNational Nature Science Foundation of China (Grant no.11202011) and the Fundamental Research Funds for theCentral Universities (Grant no. YWK13HK11). The authorsfully appreciate the financial supports. The authors wouldalso like to thank the reviewers and the editor for their manysuggestions that helped improve this paper.
References
[1] R. Lu, Y. Xu, and A. Xue, βπ»β
filtering for singular systemswith communication delays,β Signal Processing, vol. 90, no. 4,pp. 1240β1248, 2010.
[2] R. Lu, H. Li, A. Xue, J. Zheng, and Q. She, βQuantizedπ»β
filtering for different communication channels,β Circuits,Systems, and Signal Processing, vol. 31, no. 2, pp. 501β519, 2012.
[3] A. Garulli, A. Vicino, and G. Zappa, βConditional centralalgorithms for worst case set-membership identification andfiltering,β IEEE Transactions on Automatic Control, vol. 45, no.1, pp. 14β23, 2000.
[4] S. F. Schmidt, βApplication of state space methods to navigationproblems,β in Advanced in Control Systems, pp. 293β340, Aca-demic Press, New York, NY, USA, 1966.
[5] D. Woodbury and J. Junkins, βOn the consider Kalman filter,βin Proceedings of the AIAA Guidance, Navigation, and ControlConference, AIAA Paper 2010-7752, Toronto, Canada, 2010.
[6] A. H. Jazwinski, Stochastic Processes and Filtering Theory,Academic Press, New York, NY, USA, 1970.
[7] B. D. Tapley, B. E. Schutz, and G. H. Born, Statistical OrbitDetermination, Academic Press, New York, NY, USA, 1st edi-tion, 2004.
[8] R. Zanetti and C. D. Souza, βRecursive implementations ofthe consider filter,β in Proceedings of the AAS Jer-Nan JuangAstrodynamics Symposium, College Station, Tex, USA, 2012.
[9] G. J. Bierman, Factorization Methods for Discrete SequentialEstimation, Dover Publications, 2006.
[10] D. P. Woodbury, M. Majji, and J. L. Junkins, βConsideringmeasurement model parameter errors in static and dynamicsystems,β Journal of the Astronautical Sciences, vol. 58, no. 3, pp.461β478, 2011.
[11] S. A. Chee and J. R. Forbes, βNorm-constrained considerKalman filtering,β Journal of Guidance, Control, and Dynamics,vol. 37, no. 6, pp. 2048β2053, 2014.
[12] Q. Xia, M. Rao, Y. Ying, and X. Shen, βAdaptive fading Kalmanfilter with an application,β Automatica, vol. 30, no. 8, pp. 1333β1338, 1994.
[13] C. Hide, T. Moore, and M. Smith, βAdaptive Kalman filteringfor low-cost INS/GPS,β Journal of Navigation, vol. 56, no. 1, pp.143β152, 2003.
[14] C. Hu,W. Chen, Y. Chen, andD. Liu, βAdaptive Kalman filteringfor vehicle navigation,β Journal of Global Positioning Systems,vol. 2, no. 1, pp. 42β47, 2003.
[15] K.H.Kim, J.G. Lee, andC.G. Park, βAdaptive two-stageKalmanfilter in the presence of unknown random bias,β InternationalJournal of Adaptive Control and Signal Processing, vol. 20, no. 7,pp. 305β319, 2006.
[16] D. H. Zhou, Y. G. Xi, and Z. J. Zhang, βA suboptimal multiplefading extended Kalman filter,β Acta Automatica Sinica, vol. 17,no. 6, pp. 689β695, 1991.
8 Mathematical Problems in Engineering
[17] Y. Geng and J. Wang, βAdaptive estimation of multiple fadingfactors in Kalman filter for navigation applications,β GPS Solu-tions, vol. 12, no. 4, pp. 273β279, 2008.
[18] W. Gao, L. Miao, and M. Ni, βMultiple fading factors kalmanfilter for sins static alignment application,β Chinese Journal ofAeronautics, vol. 24, no. 4, pp. 476β483, 2011.
[19] H. E. Soken and C. Hajiyev, βAdaptive unscented Kalmanfilter with multiple fading factors for pico satellite attitudeestimation,β in Proceedings of the 4th International Conferenceon Recent Advances in Space Technologies (RAST β09), pp. 541β546, Istanbul, Turkey, June 2009.
[20] D. L. Snyder, βInformation processing for observed jumpprocesses,β Information and Computation, vol. 22, no. 1, pp. 69β78, 1973.
[21] L. Ozbek and F. A. Aliev, βComments on adaptive fadingKalman filter with an application,β Automatica, vol. 34, no. 12,pp. 1663β1664, 1998.
[22] J. Deyst and C. F. Price, βConditions for asymptotic stability ofthe discrete minimum-variance linear estimator,β IEEE Trans-actions on Automatic Control, vol. 13, pp. 702β705, 1968.
[23] H. W. Sorenson and J. E. Sacks, βRecursive fading memoryfiltering,β Information Sciences, vol. 3, no. 2, pp. 101β119, 1971.
[24] W.Qian, L.Wang, andY. Sun, βImproved robust stability criteriafor uncertain systems with time-varying delay,βAsian Journal ofControl, vol. 13, no. 6, pp. 1043β1050, 2011.
[25] J. L. Crassidis and J. L. Junkins, Optimal Estimation of DynamicSystems, Chapman & Hall/CRC Press, Boca Raton, Fla, USA,2nd edition, 2012.
[26] P. S. Kim, βSeparate-bias estimation scheme with diverselybehaved biases,β IEEE Transactions on Aerospace and ElectronicSystems, vol. 38, no. 1, pp. 333β339, 2002.
Submit your manuscripts athttp://www.hindawi.com
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttp://www.hindawi.com
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
CombinatoricsHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
International Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com
Volume 2014 Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Stochastic AnalysisInternational Journal of