ѧ
JOURNAL OF ASTRONAUTICS
1999 2ڡNo.2 1999



гҰֵͳϼ
֡
ժҪҰֵѧԶ,Ұֵķģ,˶Ұֵ㼰ҰֵߵġȽ-Ȼȡ鼼,˷㡣
ʡͳϡҰֵģ
STATISTICAL DIAGNOSIS METHOD FOR 
OUTLIERS FROM SPACECRAFT TRACKING DATA
Hu Shaolin Sun Guoji
(System Engineering Institute,Xi'an Jiaotong University.Xi'an.710049)
AbstractA modified definition and some classified models are built for outliers,a high efficient robust-LR (LR:likelihood ratio) testmethod is established to detect and to diagnose multi-outliers or patch-types outliers from measurement data set,and a simulation is done to show the availability of this new diagnosis approach.
Key wordsStatistical diagnosis Outliers Modeling
1
֪,ǰгݸ߾ȴй㷺õĶʽ˲ƽ΢ּ׷,ǻKalman˲۵ʵʱԤͿƼ,ӦͳѧĽǶȶԿΪ:һĹʽģͼٶ,ݶԶ̬ĿһٲγɵݼD,ͳƵĻźŴķƻԤĿ˶״̬ԡɴ˿ɼ,ݼDǽͳƶϻ򹤳̾ߵ,Dиݵ,ֱӰоľȷȺͿɿԡ
гݴʵ,ʹǸ߾ȵĲ豸,ڶżȻصۺӰ,ݼ12%ʱ1020%(,״иǸ)ݵƫĿֵݴⲿ쳣ΪҰֵ
ҰֵԸ߾ݴʮֲӰ졣ʮ,ͳƽо,ǻС۵ģͲŹơ˲㷨(ʽ˲Kalman˲)Ƶ׷ļ׹,ԲаҰֵ㷴ӦΪУ1,2,5,7,8ݡݴʵҲ֤ʵ,ںгº,ݺԼԤҪһԸĽľȡߴ൱Ҫԡ
Ұֵʶ𷽷ʹһ漰ѧ֪ʶۺõӦо⡣ں칤и೤βݼ,Ҫ׼ȷʶеȫҰֵһʮֲ׵¡رŹ̶ӻ͸ټѸռ,ݼԽԽάԽظ,ҰֵĹԽѡ,ЧɿҰֵϷ,ҪΪݴӡ
,Ұֵģͷ,һ׼Ұֵͳ㷨
2ҰֵĽģ
ʲôҰֵ?̽ݷݴ,Ұֳֵ쳣ֵ(Outliers)䶨һֱȷ,ֲ̬ͬȵӦͳѧҶ䶨Ҳһ5ݡģ5жԼֳõĶйΪ꾡ͱȽϡ
ĽҰֵĶΪ:Dƫ󲿷ƵСݵ㡣һǿֵġơ,ƫݼı仯Ϊб쳣,ȷָҰֵڼDֻռС(಻һ),ֱǺġ岻ԸǼ,Edeworth(1887)Ķģ6,7ݵĶ,ҲʱмϵͳеҰֵ,нΪ㷺Ӧԡ
2.1Ұֵĳ
ݼгҰֵԭܶࡣͺгݶ,ҰֵҪ¼ԭ:
(1)ͼ¼ʱĹʧ,Լݸƺͼ㴦ʱֵĹʧԴɴ˲Ϊʧ(Gross Error)
(2)ı仯ȡĸͻȻıʹòԭģͲ,زʱͻĳ֡״ʱӦ״̬ĲȶгʱĿıػ㸽ȡ,ȵȡ
(3)ʵʲҲܳһ쳣,ȲԲʹĹʧҲͻǿӰصµ,ĳЩӳβֲ(long-tailed distribution)(,tֲ)õĽ
2.2Ұֵķ
̲гֵҰֵ,Ƚϳ¼:
(1)Ⱥ㡡ֳ,ͳоҲȽϳһ7ݡ,ҰֵҪΪȺ㡣
(2)ҰֵĻص,ĳһʱtiĲǷΪҰֵti-1ti+1ʱݵޱȻϵ,Ƚϳǵtiʱ̵Ĳݳ쳣ʱ,tiʱ̵һ(ʱ)Ǻõ,Ұֵĳǹġĸ岻,̬й(isolated)쳣ֵĳҲǱȽձ֮һ
(3)Ұֵߵ㡡ưߵ(Patchy),ָƬֵ쳣ݡĻ,ti-pʱֵ̳Ұֵ,Ҳܴy(ti-p+1),,y(ti)ƫֵ״ٸĿĲ,ҰֵߵĳǱȽϳĹ
2.3Ұֵģ
ٶĳվԿռĿиٹ۲,ǲݼΪ
D={y(t1),y(tN)}(1)
(1)ݵġ(additive)ֽ:Ŀ˶,ǿԼٶy(t)ɺطֽĲ:
y(t)=ytr(t)+yp(t)+s(t)+0(t)(2)
Уytr(t)ΪƷ,Ŀ˶;
yp(t)Ϊڷ,ĿԱ仯,Ŀԡͬʱ,Ա仯,ÿһʱ˶״һ;
ۦs(t)ǲ,ӳǲ豸״̬ĿлжֲȷŶͬõĽ
ܦ0(t)Ϊ̵ȾͻԷ,ýʹݷƫ,ҪôΪ,Ҫô׼,
(3)
ʽ,Ϊ쳣ݳֵļʡݷ,ͨ1%2.5%֮䡣
(2)쳣(Additive Outliers,AO)ģ:ĳЩʱti0,ڲͻȻǿӰӰ,Լ¼ʧ,ʹòƫ󲿷ֵ,ģΪ:
0(t)=˦(i,i0)(4)
ʽ,Ϊ쳣,(i,i0)ΪKronecker delta
(3)Ϣ쳣(Innovation Outliers,IO)ģͣ4:ڹ۲ֵģ(2),̦0(t)ӳβֲҾйƽʱ,ƽִ̼ʱз,ɽһٶ0(t)ARMA(p,q)ģ͡
3Ұֵͳ
ڼҰֵʶ,ף1,6,7ݼοжϸ;Ұֵ,ҲѾдоɹɹã4;ڶҰֵ(رǰߵҰֵ)봦,漰ǰ,׵ĳɹĴв
ǵߵҰֵͳϵļѶȺ͹ʵüֵ,ĽԴ˽һЩ̽оĵĻ뷨,βݼҰֵϺʹֳ׶:
ݴ:һȽġġ;
ڿɵɸѡ:ͨʵáġߵıȽ,ȷܰҰֵsλ(ti1,,tis);
ͳ:ͳͼ,һʵȷsȷʵΪ쳣ݡ
,ӹӦõĽǶȽʵһĻ輰㷨
3.1ġȽʽ
ݷ˶ѧ͹ѧ,Լٶ:޳ȵı۲⻡,Ŀ˶켣y(t)Ǽι⻬ġWeierstrassƽ,y(t)ԱʽǶʽһ±ƽ,
(5)
,TΪԱ仯ype(t)ݱڡ
ʽ(1)ĲݼD={y(ti):i=1,2,,N},ģ(5)ԸдΪԻعģ͵ʽ:
(6)
ʽУ1)
2)=(a0,a1,,a1,b1,bm,c1,,cn),(t)=s(t)+0(t)
ģͲ¼ҵM-ȽƷ:
(7)
У1)
2)(c1c2ΪʵѡȡķǸ)
Ȼ,ѡȡc1=+ʱ,()=0.52ͦ()=ؾΪ޽纯,ʽ(7)ΪС˹Ʒ;ʵѡȡc1c2Ϊ޷Ǹʱ,()ͦ()Ϊн纯,ʱʽ(7)ȷĲ(,)ΪȽ-Թ,õĿҰֵ
,ģͲM-͹Ʒ(7)һ㷨:
(8)
óֵΪ:
(9)
Ƚ-Źֵ,Ѹy(t)ľпҰֵȽ:
(10)
ע:ڹԤʱ,Ϊ˼򻯼,ҲʵʽHuberͻHampelͦպ:
(1)(11)
(2)(12)
ʽ,cΪݺ;ʵѡȡķǸ
3.2Ұֵλõȷ
ٲݼDиݵĹ淶ϲв:
(13)
ϲвɢͼ,¼ʹʱ㼰ݸsá3׼򡱿ɳΪsпҰֵ㡣ɴ˵óΪҰֵݵs,ڵλ{ti1,,tis}
3.3ҰֵȻȼ
ٶģ(2)Ϊֲͬ(iid)GaussݼDвҰֵʱ,֤7:ڿɵ{y(ti1),,y(tis)}ǷΪҰֵȻͳ
(14)
ʽ,RssRssֱΪɾɵ{y(ti1),,y(tis)}ǰ/СϵĲвƽ͡
ʽ(14)Կ,ȻȺLR(s)ǵϸ,̬Լٶ¿֤7ͳӲΪĦ-ֲ,ͳ:
(15)
ΪݺԼͳƼָꡣ
-ֲĦ·λΪC,(=0.050.025)ͳLR(s)1-Cʱ,100(1-)%ĸΪDпɵ{y(ti1),,y(tis)}Ĺݡ
4
ͼ1Ǽɵĳػ״ٿռĿꡰ20,ƵΪÿ10㡣ͼ1е5091110͵200Ϊⲿӷγɵ22Ұֵ,ѧģΪ4״ʽ,ģ趨(1);Ұֵƫ:0(50)=-10,0(150)=10;ߵҰֵѧģΪ:0(i+1)=1-0(i),0(90)=-5(i=90,,109)
,±1ͬʱģϵ(ȫ)С˹ơʽ(8)(9)M-Ƚƺ(޵)С˹,ԼϵĲвƽ͡

ͼ122Ұֵɢͼͼ2M-͹Ƶߡ
1ģͲ趨ֵͳƹֵ

a0a1a2a3a4Rss
趨-5.016.0-0.5-0.030.0010.1
ȫС-4.85715.960-0.5157-0.027630.0009241.83099653.746
M-͹-4.93015.984-0.4998-0.029910.0009960.101392.00452
޵С-4.94115.993-0.5017-0.029780.0009930.099741.71095

ֱM-͹ƺȫС˹Ƽ㲻ͬϵı׼в,ͼ3ͼ4ʾ:

ͼ3M-͹Ƶı׼ϲвͼ4ȫС˹Ƶϲв
ӱ1ͼ2Կ,M-͹ƾõĿҰֵӰ;ͼ3Կ,ϲввģ͵Ʒ,趨22ҰֵӦĲвֵԶԶ3,вľֵԵС3,3׼󱨿ɵĸΪ0;ͼ4Կ,ҰֵСϵĲӰʮԵ,ввԵƷ,22Ұֵ⻹11Ĳвֵ3,3׼򡱵󱨿ɵ11
˵,ĵġȽ-Ȼȡ鷽ҰֵߵнϸߵļЧʺͿɿԡ
л:еõԺʿָͰ,ڴ˽л
ߵλͨѧϵͳо..710049
ο
1PJ.Huber FJ.Robust Statistics.Wiley,New York,1981
2Hampel,et.Robust Statistics-approach based on influence.John Wildy and Sons Press,1986
3,.ԲǶʽ˲Ԥ.ͨѧѧ,1996,30(2):110-116
4Tiao GC.Autoregressive Moving Average Models.Inter.Problems and Outlier Detection in Time Series,Handbook of Stat.5
5,.̬ݿо.ͳӦø,1996,11(3):244-248
6Wang Xueren,et.Assessment of Local Inference in Multivariate Analysis.Technical Reports(1990-1991),Yunnal University Press,1991
7Τɵ.ͳ.Ͼ:ϴѧ,1991
8,.̬ϵͳнӰ˲.ۼӦ,1992,11(1):36-45
ΪͨѧоʿѧλĻ(DFXJU1998-10)
ո:1997421,޻:1998224
