1、)()(),1()1(nnnnn1vxx)()()()(nnnn2vxHz(6.1.1)状态方状态方程程(6.1.2)测量方程测量方程为为M维系统状态矢量维系统状态矢量为为MM维转移矩阵维转移矩阵为为N维测量矢量维测量矢量为为NM维测量矩阵维测量矩阵)(nx),1(nn)(nz)(nH0v2)(nEknknknE0Qvv1T11)()(knknknE0Qvv2T22)()(0vvT21)()(knE0v1)(nE为为M维系统噪音矢量(白噪音矢量)维系统噪音矢量(白噪音矢量)。为为N维测量噪音矢量(白噪音矢量维测量噪音矢量(白噪音矢量))(n1v)(n2v)()(),1()1(nnnnn1vxx
2、)()()()(nnnn2vxHz(6.1.1)状态方状态方程程(6.1.2)测量方程测量方程图图6.1系统的动态模型系统的动态模型)()(),1()1(nnnnn1vxx)()()()(nnnn2vxHz(6.1.1)状态方程状态方程(6.1.2)测量方程测量方程卡尔曼滤波就是对由式卡尔曼滤波就是对由式(6.1.1)(6.1.6)所描述的系统所描述的系统(、为已知),为已知),根据测量矢量根据测量矢量 对状态矢量对状态矢量 进行估计,进行估计,使估计误差的均方差为最小。使估计误差的均方差为最小。H1Q2Q)(nz)(nx预测。根据测量值预测。根据测量值 ,估计,估计 。滤波。根据测量值滤波。
3、根据测量值 ,估计估计 。)1(z)1(nz)(nx)1(z)1(nz)(nz)(nx 两座标引导雷达两座标引导雷达Melbourne Area Approach Control Centre)()()1(nvnn)()()1(nTnn)()()1(nvnunuu)()()1(nuTnrnr信号处理机每隔信号处理机每隔T秒输出飞机的一组径向距离秒输出飞机的一组径向距离r和方位和方位的数据,但含有噪音。的数据,但含有噪音。录取设备要对这些数据进行处理,抑制噪音并建立起飞机的航行轨迹(航迹)。录取设备要对这些数据进行处理,抑制噪音并建立起飞机的航行轨迹(航迹)。雷达输出数据的周期雷达输出数据的周期
4、T通常为秒量级,在此时间可假定飞机作匀速运动。通常为秒量级,在此时间可假定飞机作匀速运动。r(n)为nT时刻飞机的径向距离;为飞机的径向速度;为飞机方位;为飞机角速度。和 为零均值白噪音,机动噪音nTtdttdrnu)()()(nnTtdttdn)()(uv(n)v(n)()TnTnnnunrn)(),(),(),()(xTunvnvn)(,0),(,0)(1v10001000010001),1(TTnn22)(nvEuu2200000000000000)()()(unnEnT111vvQ22()E v n)()(),1()1(nnnnn1vxx状态方程状态方程(6.1.9)、)()()1(n
5、vnn)()()1(nTnn)()()1(nvnunuu)()()1(nuTnrnr)()()(21nznznz01000001)(nH)()()(nnnnnr2v)()()(2nnnnz22)(nnErr2200)()()(rnnEnT222vvQ)()()()(nnnn2vxHz测量方程测量方程(6.1.14)22)(nnE信号处理机每信号处理机每T秒送一次有误差的秒送一次有误差的 、)(nr)(n)()()(1nnnrnzrTnnnunrn)(),(),(),()(x)1,()1,()1,()(2nnnnEnnEneeeT)1,()1,()1,(nnnnEnnTeeP)1,()()1,(
6、nnnnnxxe预测在于根据测量值 ,估计 ,估计值记为 )1(z)1(nz)(nx)1,(nnx 预测误差的均方差即均方误差(纯量)预测误差相关阵(矩阵)2)1,()1,()(nnEnMinnne X卡尔曼预测即最佳卡尔曼预测估计 预测误差矢量6.1.2预测预测(,1)()(,1)n nnn nzHx)1,()1,()()1,(nnnnnnn2vxHz)()()1,()()(nnnnnn2TQHPHR)()1,()()(nnnnn2veH)()1,()()()(nnnnnn2vxxH)()()(nnEnTR)1,()()()1,()()(nnnnnnnnxHzzz)()(),1()1(nnn
7、nn1vxx)()()()(nnnn2vxHz(6.1.1)状态方程状态方程(6.1.2)测量方程测量方程状态矢量预测值为测量矢量之预测值为)1,(nnx新息矢量 新息相关阵)1,()1,()1,(nnnnEnnTeeP)1,()()1,(nnnnnxxeknknknE0Qvv1T11)()(knknknE0Qvv2T22)()(0vvT21)()(knE)1,()()()()1,(),1(),1(nnnnnnnnnnnxHzKxx)()()1,(),1(),1(nnnnnnnnKxx最佳的卡尔曼预测滤波器递推方程最佳的卡尔曼预测滤波器递推方程 图 6.2 卡尔曼预测框图(6.1.27)(,1
8、)n nx1TQPP),1()(),1(),1(nnnnnnn1()(1,)(,1)()()(1,)(,1)()()(,1)()nnnn nnnnnn nnnn nnT1TT2KPHRPHHPHQ)1,()()()1,()1,()(nnnnnnnnnPHKPP),1()1,(nnnn1图图 6.3 卡尔曼预测卡尔曼预测及滤波框图及滤波框图)(nK增益矩阵 递推方程)1,(nnP(6.1.28)(6.1.29)(6.1.30)(6.1.31)预测误差相关阵 递推方程(Riccati方程方程)预测预测(6.1.27)(6.1.28)(6.1.29-31))1,()()()()1,(),1(),1(
9、nnnnnnnnnnnxHzKxx最佳的卡尔曼预测滤波器递推方程最佳的卡尔曼预测滤波器递推方程),1()1,()(nnnnnxx),1(),1()(nnnnnxx1)(),1(),1(nnnnnxx滤波指根据测量值 ,估计 。6.1.3 滤波滤波)1(z)1(nz)(nz)(nx记为)(),(nnnxx6.1.4 初始条件和卡尔曼预测算法流程初始条件和卡尔曼预测算法流程初始条件初始条件0TPxxP)1()1()0,1(E0 x)0,1(0PxxxxPTEEE)1()1()1()1()0,1()1()0,1(xxE表表6.1 卡尔曼预测算法流程(卡尔曼预测算法流程(1)模型:)()(),1()1
10、(nnnnn1vxx)()()()(nnnn2vxHz)(n1v和)(n2v为零均值白噪音,2,QQ1),1(nn),(nH2,QQ1为已知),1()1,(nnnn1其相关阵分别为、初始值:)1()0,1(xxE0PxxxxPTEEE)1()1()1()1()0,1(输入:)(,),1(),1(nzzz计算:对n=1,2,(1)本次增益及新息1)()1,()()()1,(),1()(2TTQHPHHPKnnnnnnnnnn)1,()()()(nnnnnxHz(2)求本次预测值)()()1,(),1(),1(nnnnnnnnKxx(3)准备下次的),1(nn P)1,()()()1,()1,()
11、(nnnnnnnnnPHKPP1TQPP),1()(),1(),1(nnnnnnn表表6.1 卡尔曼预测算法流程(卡尔曼预测算法流程(2))()()()(nvknhkdnxk图6.4 自适应滤波器及其输入信号产生模型)()()(nnnyTwx输入信号自适应滤波器2w1wMw)()(ndnExdxr)()(nnETxxxxRxdxxoptrRw1)(2neE()()()()()()Te nd ny nd nnnxw)()(2min2minneEneEoptoptToptnndnewx)()()()()()(nenndoptoptTwx状态变量状态变量2w1wMw)(noptw)()()()(ne
12、nnndoptoptTwx)1()(nnoptoptww状态方程状态方程测量方程测量方程)(noptw测量矢量为纯量)(nd系统噪音矢量为0)(neopt系统转移矩阵)1,(nn为单位阵I测量矩阵)(nH为矢量)(nTx)(neopt为零均值,方差为min的白噪音 测量噪音矢量为纯量状态矢量为)()()()(nennndoptoptTwx()(1)(1)()optoptoptoptnnnnwwww状态方程状态方程测量方程测量方程(,1)n nI()n 1v0()()TnnHx()()nd nz2()()optnenv1Q02min()nQ)()(),1()1(nnnnn1vxx)()()()(
13、nnnn2vxHz(1,)(1,)()(,1)(,1)(1)(1,)()(,1)(1)optoptoptoptnnnnnn nn nnnnnn nnxwwxwwPPPP状态变量估计状态变量估计1TQPP),1()(),1(),1(nnnnnnn1()(1,)(,1)()()(,1)()nnnn nnnn nnTT2KPHHPHQ)1,()()()1,()1,()(nnnnnnnnnPHKPP)1,()()()()1,(),1(),1(nnnnnnnnnnnxHzKxx(,1)n nI()n 1v0()()TnnHx()()nd nz2()()optnenv1Q02min()nQ)1()()()
14、(nnnnTPxkIP)1()()()1()(nnnnnTPxkPP1min)()1()()()1()(nnnnnnTxPxxPk)1()()()()1()(nnndnnnoptToptoptwxkww)()()()1()(ndndnnnoptoptkww)1()()(nnndoptTwxTopt,0,0,1)0(wTopt,0,0,1,0,0)0(wIP)0(min)1()()()()1()(nnnnnnTPxxPPPmin)()()(nnnxPkmin)()()(nnnkxPmin)()()1()()(nnnnnTkxPxkI IRP1min)(nnxx11min)(xxnnRP nkxx
15、Tkxkxn1)()(1R)()()()(11nkxkxnnkTxknkTkkn11min)()()(xxPnkTkkn111minmin)()()0()(xxPPmin11)()()1()(nnnnTxxPP1)(MnnTxxrRP)1()(1minMnnMM(1)M MMI2/1),1(nn0Q 1)()(nnTuH12QknknkvnvE01)()()()()()(nvnnnzTxu)()1(2/1nnxx(6.3.1)(6.3.2)为正实数为正实数(1)M()MM(1 1)(1)M(1)M 1)()1,()()(nnnnnrTuPu()()()(,1)()()(,1)()TTnz nn
16、n nnnn nv nuxuxx新息矢量为纯量)(n的相关矩阵)(nR为纯量)1,()()()(nnnnnxHz)()()()(nvnnnzTxu)()1(2/1nnxxI2/1),1(nn)()(nnTuH)()()1,()()(nnnnnn2TQHPHR)()()(nnEnTR)1,()1,()1,(nnnnEnnTeeP)1,()()1,(nnnnnxxe预测误差相关阵预测误差矢量)()1,()(1)()1,()()()1,()(2/112/1nnnnnnnnrnnnnTuPuuPuPk增益矩阵为矢量,记为)(nk1)()1,()()()1,(),1()(2TTQHPHHPKnnnnnn
17、nnnn)()()()(nvnnnzTxu)()1(2/1nnxxI2/1),1(nn)()(nnTuH)1,()()()1,()(2/1nnnnnnnTPukPP)(),1(1nnnPP)1()1,(1nnnPP)1,()()()()1,(),1(2/1nnnnznnnnnTxukxx(1,)nnP1/2求本次预测值求本次预测值-基本递推方程基本递推方程(6.1.27)准备下次的准备下次的预测误差相关阵预测误差相关阵 (6.1.28-29),1(nn P)1,()()()1,()1,()(nnnnnnnnnPHKPP1TQPP),1()(),1(),1(nnnnnnn)()()()(nvnn
18、nzTxu)()1(2/1nnxxI2/1),1(nn)()(nnTuH)1,()()()()1,(),1(),1(nnnnnnnnnnnxHzKxx)1,()()()()1,(),1(2/1nnnnznnnnnTxukxx基本递推方程(6.3.11)(),1(1nnnPP)1,()()()1,()(2/1nnnnnnnTPukPP1/2(,1)()()1()(,1)()Tn nnnnn nnPukuPu增益矢量(6.3.10)预测误差相关阵(6.3.12)(1,)nnP1/21(,1)(1)n nnPP)1()()()1()(2/11nnnnnTPukPP)()1()(1)()1()(2/1
19、nnnnnnTuPuuPk)1()()()1()(2/11nnnnnTPukPP)()1()(1)()1()(2/1nnnnnnTuPuuPk)(),1()(nnnnPPP)(),1(nnnPP)(),1(1nnnPP)1,()()()1,()(2/1nnnnnnnTPukPP1/2(,1)()()1()(,1)()Tn nnnnn nnPukuPu1(,1)(1)n nnPP)1()1,(nnnPP)(),1(1nnnPP1/2(,1)()()1()(,1)()Tn nnnnn nnPukuPu表表6.2 方差卡尔曼滤波算法方差卡尔曼滤波算法系统模型)()1(2/1nnxx)()()()(n
20、vnnnzTxu已知:)(nTu1)(2nvE输入测量值(纯量):)(,),2(),1(nzzz初始值:)1()0,1(xxE0)1()1()1()1()0(PxxxxPTEEE计算:,3,2,1n)()1()(1)()1()(2/1nnnnnnTuPuuPk)1,()()()()1,(),1(2/1nnnnznnnnnTxukxx)1()()()1()(2/11nnnnnTPukPP表表5.1 递推最小二乘(递推最小二乘(RLS)算法流程)算法流程 0 xw)0()0(,)0(1IC为小的正实数,2,1n(1)取得)(nd,)(nx (2)更新增益矢量 )()1()()(nnnnHxCx)(
21、)()1()(nnnnxCg(3)更新滤波器参量 ()(1)()()()(1)Tnnn d nnnwwgxw(4)更新逆矩阵1()(1)()()(1)TnnnnnCCgxC初始条件:运算:对表表 6.3 Kalman(表(表6.2)与与 RLS(表(表5.1)对应表对应表(1)Kalman RLS名称 变量测量信号)(nz)(2/ndn 需要信号转移矩阵 预测权矢量 增益矢量(6.3.10)预测误差相关矩阵(6.3.14)(nTu),1(nn x)(nk)(nP)(nTx)(2/)1(nnw)(2/1ng)(1nC)(1n1xR输入矢量转置 滤波权 增益矢量(5.1.30)输入矢量相关矩阵之逆
22、(5.1.29)变量 名称新息(6.3.8)新息均方值(6.3.9)初始条件)(n)(1nr)0,1(x)0(P)(2/nen先)(n)0(w)0(1C 先验误差(5.1.37)变换系数(5.1.42)表表 6.3 Kalman(表(表6.2)与与 RLS(表(表5.1)对应表对应表(2)名称 变量变量 名称初始条件)1()()()()1()1()(11nnnrnnnntPuuPPP)()()1()()()1,()(12/112/1nrnnnrnnnnuPuPk1)()1()(1)()1,()()(nnnnnnnnrTTduPuuPu)()()(2/2/1nnnTPPP)1()()1(1)(2
23、/12/nnnnTTTPuP0A)1()1()(1)(2/12/12/1nnnnTP0PuA)()()(nnnTAAM)1()1()1(2/2/1nnnTPPP)1()()1()1()(1)()1()()(2/12/12/1nnnnnnnnnTTPuPPuuPuM)1()()()()1()1()(11nnnrnnnnTPuuPPP)()()1()(12/1nrnnnuPk1)()1()()(nnnnrTuPu)1()()1(1)1()1(1)(2/2/12/2/12/12/1nnnnnnTTTTPuP0P0PuM)()()()()1()1()(12221112/12/12/1nnnbnnnnT
24、TBb0QP0Pu)()()()(222111nnnbnTBb0BBAQ)()()()()()(222111222111nnnbnnnbTTTB0bBb0)()(2/122nnPB)()()(2/12/121nrnnkb)()()1()(2/12/121nrnnnuPb)()(2/111nrnb)1()()()()(2/122222121nnnnnTTPBBbb)()1()()(2/11121nnnbnuPb)(1)()1()()(211nrnnnnbTuPu)()()1,(),1(2/1nnnnnnkxx)1,()()()(nnnnznTxu12/12/1)()()()(nrnrnnkk)(
25、)()()()()1()1()(12/12/12/12/12/12/1nnrnnrnnnnTTPk0QP0Pu22211122211211222112110bbbbbbbcsscaaaa121111sacab222121sacab222122casab0121112casab21221112aaasIQQ100111T122 sc)1()()()()1()(nnndnnnTwxgww)1()()()()1()(2/2/12/2/12/)1(1nnndnnnTnnnwxgww)()()()(2/112/1nnnngg)()()()()()()()(12/12/12/12/12/12/12/1nn
26、nnnnnnTTCg0QC0Cx)()()(2/12/1nnnTPPP)()()()()()(),1(0)()(1)()1()1,()()1(2/12/12/12/12/2/2/2/12/2/1nrnnnnrnnnnnnznnnnnTTTTTTTTPuPxPQ0PxuP)()()(2/1nnnxTTRpw)()()()()()()()(1)()1()()1(2/12/2/12/12/12/12/1nnnnnennnndnnnTxTTxTTxRxp0RQ0pxR先)()()()()(2/12/nnnnnxdxTxrRwRp)()()(2/2/1nnnTxxxRRR)()()(nnnxdxrwR)
27、()()(2/1nnnTxTPRw表 6.4 基于 的平方根卡尔曼算法流程 系统模型)()1(2/1nnxx)()()()(nvnnnzTxu已知:)(nTu1)(2nvE输入测量值(纯量):)(,),2(),1(nzzz初始值:)1()0,1(xxE计算:,3,2,1n)(2/1nP2/102/12/1)1()1()1()1()0(PxxxxPTEEE)()()()()()1()1()(12/12/12/12/12/12/1nnrnnrnnnnTTPk0QP0Pu 12/12/1)()()()(nrnrnnkk)1,()()()(nnnnznTxu)()()1,(),1(2/1nnnnnnk
28、xx表6.5 基于 递推的平方根RLS算法)(2/1nR初始条件:0 xw)0()0(IC12/10()运算:对,3,2,1n(1)取得)(nd)(nx(2)计算)(2/1nC)(2/1n)()(2/1nng)()()()()()()()(12/12/12/12/12/12/12/1nnnnnnnnTTCg0QC0Cx(3)计算)(ng)()()()(2/112/1nnnngg(4)计算)(nw)1()()()()1()(nnndnnnTwxgww表 6.6 基于 的平方根卡尔曼算法流程 系统模型)()1(2/1nnxx)()()()(nvnnnzTxu已知:)(nTu1)(2nvE输入测量值
29、(纯量):)(,),2(),1(nzzz初始值:)1()0,1(xxE计算:,3,2,1n)(2/1nP2/102/1)0(PP)()()()()()(),1(0)()(1)()1()1,()()1(2/12/12/12/12/2/2/2/12/2/1nrnnnnrnnnnnnznnnnnTTTTTTTPuPxPQ0PxuP12/2/)()(),1(),1(nnnnnnTTTTPPxx表6.7 基于 递推的平方根RLS算法初始条件:运算:对,3,2,1n(1)取得)(nd)(nx(2)计算)(2/1nxRIR)0(2/1x0w)0(0p)0()()()()()()()()(1)()1()()1(2/12/2/12/12/12/12/1nnnnnennnndnnnTxTTxTTxRxp0RQ0pxR先)()()(2/1nnnTTRpw