Զѧ
ACTA AUTOMATICA SINICA
1999ꡡ255ڡVol.25 No.5 1999



Kalman˲ĿҰֵ1)
塡
ժҪ˶̬ϵͳ쳣ݶKalman˲ĲӰ.һݴ˲㷨.˲㷨ȿԳϢȷ˲ľȣֿЧ쳣ϢĲӰ˲ƿɿԣﵽԹݵݴ.֤ͨ˸㷨Ч.
ؼʡKalman˲̬ϵͳҰֵݴ˲.
FAULT-TOLERANT IMPROVEMENT ON KALMAN FILTER
HU FengSUN Guoji
(System Engineering Institute,Xi'an Jiaotong University,Xi'an710049)
AbstractThe influence of outliers on the Kalman filter is analysed,and a new set of modified algorithms,which are fault-tolerant to outliers is built. These new algorithms can not only sufficiently absorb good innovations to increase filtering precision but also efficiently restrain bad influence coming from outliers.These new filters have high security and fault-tolerant capability when abrupt faults take place.Simulation results show that the modified filtering algorithms are practical and efficiently resistant to outliers in sampling data.
Key wordsKalman filter, dynamic-measurement system, outlier, fault-tolerant filter.
1
60״̬ռ۷չKalman˲һԵƹϵϵͳ״̬˲㷨߱ѧʣ1ݣơźŴ̼غ͹ϣ2ݵʮֹ㷺Ӧ.
ǣʮоʾ36ݣKalman˲㷨ȱԲݹʧĿԺͶԴͻԹϵݴ.ӦKalman˲㷨ʵľҲ֤ʵмʹĹʧKalman˲ЧҲصĲӰ죬Ŀ״̬Եƫϵͳʵ״̬.
гƴͻԹʱĲϢ򺬹ʧĲΪҰֵ(Outliers).Ĵӹ̿ƺźŴʵʳϵͳҰֵKalman˲ĲӰ죬һпҰֵɿ͵㷨.
2ҰֵKalman˲Ӱ
Զ̬ϵͳ
(1)
״̬Kalman˲㷨Ϊ1
(2)
ʽϢ,˲Kk+1=k+1Hk+1Hk+1k+1Hk+1+R(k+1)-1;k+1=Ak+1(I-KkHk)kAk+1+R(k),R(k)=cov(k,k),R(k)=cov(k,k).
ʽ(2)Կyk+1Ϣ(innovations)ϵķʽKalman˲ֵӰ.ˣyk+1ΪֵʱϢtk+1ʱ̵˲ֵȷ(Ϊ˲Kk+1)yk+1Ϊ쳣ֵ(Ұֵ)ʱ쳣ϢҲKk+1ضϵͳ״̬Ԥдʹ˲ֵƫ.
3ҰֵеĴԲ
ᴦʵʱعУԴʵݽҰֵͳǱȽѵ.ˣΪKalman˲ݴбҪ˲㷨ݴĽʵҪ
1)˲㷨мĵƹϵڹӦã
2)۲кҰֵʱЧ˷䲻Ӱ죬ܽӰĳԤȸڣȷ˲ֵܽӽϵͳʵ״̬
3)ݼвҰֵʱܳЧϢ˲.
룬ĹһKalman˲㷨
(3)
ʽЦk+1(.)Ϊʵѡȡİι⻬ĺΪѹӰ캯Kk+1Ϊ˲棬Gk+1ΪȨ.
Ȼʽ(3)һ˲壬Ŧk+1(.)ȡĲͬԵõвͬͳʵ˲㷨.
֤ѡȡk+1(.)=1,Gk+1=I(λ)ʱʽ(3)ļΪKalman˲.ǣУk+1(.)=1õֱЧͬȿϢ쳣Ϣ˲㷨ȱҰֵŵ.
ڴˣһֺݴͨʵѡȡk+1(.)ʹõrk+1ʱk+1(.)Сӽ.
4Ѧk(.)ѡȡ
ڵ3ڵķΪҰֵݶ״̬˲ĲӰ죬ͨʵѡȡѹӰ캯k+1(.)˲
(4)
ĴС.ʵ޳ck+1ͨѡȡѹӰ캯k+1(.)ʹʽ
(5)
ʽеֵ.
ѿ(5)Ħk+1(.)кܶ࣬γɼS
(6)
棬۴ӼSѡȡѹӰ캯Цk(.)ķ.
.ϵͳ(1)ĻУx0,0,1,,1,2,Ϊֲ̬һʱѡȡȨGk+1=Hk+1k+1Hk+1+R(k+1)ںSʹ˲СĦΪ
(7)
CkΪʵѡȡ޳.
֤.ʽ(3)(2)
(8)
תΪڼSļС.ʽ(2)(3)
(9)
SĶ֪ʽ(7)ģk(.)Sʹʽ(9)ﵽС.
֤.
ʵӦУΪ򻯼Ҳɲʽģk(r)
(10)
ʽC*1ΪʵѡȡķǸ.
棬޳Ck+1ѡȡһ.
£֤(k+1k)ӱ׼ĶԪֲ̬.,rk+1=(k+1k)2ɶΪmĦ-ֲ.ȡæ-ֲŶΪ(1-)100%ϷλΪcm)(ȡ0.050.025)޳Ck+1ĺȡֵɲʽ
Ck+1=(k+1c(m))1/2(11)
ڹľӦУCk+1Ҳƾѡȡ.
5
趨ϵͳģ(1)ϵЭֱΪ

ѡȡϵͳĳ̬Ϊx(00)=(1.31.52.3),(00)=0.01I3I3ʾs׵λ.
Monte Carlo100桰(ΪS).Sе506075ֱƫ(y1,y2)=(-1)i(100,-5)(i=50,60,75)γɺ3ҰֵݼS*.S*еݣֱʽ(2)(3)״̬˲ƣͼ1ͼ2ʾ


ͼ13ά״̬ռKalman˲


ͼ23ά״̬ռݴ˲
ͼ12ԿҰֵKalman˲ʮԵĲӰ죬ݴ㷨ܹЧؿ˷ҰֵĲӰ죬˲ƵĿɿ.
лõܹʦԺʿͷǽڵָл.
1)ͨѧоԺʿѧλĻ(DFXJU1998-10)Ŀ.
߼飺塡У19641989˶ʿҵ.ΪͨѧϵͳоʿоǲĳҸΡ߼ʦйѧԱѧί.еп7μӴͺ30Σо50ƪ(11ƪĻ)ίƼȽ6.ҪоΪϵͳģ桢ϵͳۡȽ-ݴ.
19361966ѧϵͳ渱ʿѧλͨѧУ͸УѧоԺԺ.Ϊʿʦȫϵͳѧḱ³ϵͳѧѧڿıί.ҪCIMSʵڷеӦоȺе863Ҹ߼ƻȹؿȫѧάίƼо50ƪ.
ߵλͨѧϵͳо710049
ο
1ѣ¾꣬.ϵͳۣͨѧ磬1987
2Isermann R.Process fault detection based on modeling and estimation methods-a survey. Automatica,198420(4)387404
3Martin C J,Mintz M.Robust filtering and prediction for linear systems with uncertain dynamics:a game-theoretic approach.IEEE Trans on Automatic Control,1983,28(9):888-896
4壬.̬ϵͳнӰ˲.Ӧã199310(1)3645
5壬.̬ϵͳԤ㷨.ѧ199516(1)9396
6壬.̬-ϵͳԼоķչ.ͳӦøʣ199611(3)244248
ոڣ1997-09-01
޸ڣ1998-06-23
