Knowledge

Separation principle

Source 📝

794: 58:
Another instance of the separation principle arises in the setting of linear stochastic systems, namely that state estimation (possibly nonlinear) together with an optimal state feedback controller designed to minimize a quadratic cost, is optimal for the stochastic control problem with output
211: 649: 403: 97: 578: 463: 518: 638: 789:{\displaystyle {\begin{bmatrix}{\dot {x}}\\{\dot {e}}\\\end{bmatrix}}={\begin{bmatrix}A-BK&BK\\0&A-LC\\\end{bmatrix}}{\begin{bmatrix}x\\e\\\end{bmatrix}}.} 308: 277: 246: 92: 71:. More generally, under suitable conditions and when the noise is a martingale (with possible jumps), again a separation principle applies and is known as the 28:, states that under some assumptions the problem of designing an optimal feedback controller for a stochastic system can be solved by designing an optimal 78:
The separation principle also holds for high gain observers used for state estimation of a class of nonlinear systems and control of quantum systems.
72: 320: 1047: 853: 68: 39:
The first instance of such a principle is in the setting of deterministic linear systems, namely that if a stable
981:
Tryphon T. Georgiou and Anders Lindquist (2013). "The Separation Principle in Stochastic Control, Redux".
529: 414: 1103: 33: 478: 64: 36:
for the system. Thus the problem can be broken into two separate parts, which facilitates the design.
586: 59:
measurements. When process and observation noise are Gaussian, the optimal solution separates into a
1098: 206:{\displaystyle {\begin{aligned}{\dot {x}}(t)&=Ax(t)+Bu(t)\\y(t)&=Cx(t)\end{aligned}}} 869:
Tyrone Duncan and Pravin Varaiya (1971). "On the solutions of a stochastic control system".
284: 253: 222: 8: 824: 1053: 1008: 990: 1057: 1043: 924: 907: 849: 800: 48: 1012: 1035: 1032:
Proceedings of the 37th IEEE Conference on Decision and Control (Cat. No.98CH36171)
1000: 948: 919: 886: 878: 52: 40: 29: 17: 81: 1092: 1027: 1004: 939:
Anders Lindquist (1973). "On Feedback Control of Linear Stochastic Systems".
60: 1039: 55:. The separation principle does not hold in general for nonlinear systems. 1083:
Computational Aspects of Linear Control (Numerical Methods and Algorithms)
1071: 804: 1028:"A separation principle for the control of a class of nonlinear systems" 980: 32:
for the state of the system, which feeds into an optimal deterministic
891: 952: 882: 51:(LTI system hereafter), then the combined observer and feedback is 44: 995: 868: 905: 82:
Proof of separation principle for deterministic LTI systems
398:{\displaystyle {\dot {\hat {x}}}=(A-LC){\hat {x}}+Bu+Ly\,} 823:. Thus the stability of the observer and feedback are 762: 705: 658: 652: 589: 532: 481: 417: 323: 287: 256: 225: 95: 968:Stochastic Control of Partially Observable Systems 788: 632: 572: 512: 457: 397: 302: 271: 240: 205: 843: 26:principle of separation of estimation and control 1090: 938: 1070:Proof can be found in this math.stackexchange 965: 643:Now we can write the closed-loop dynamics as 1025: 862: 310:represents the internal state of the system. 908:"Information states for stochastic systems" 899: 73:separation principle in stochastic control 994: 959: 923: 890: 846:Introduction to Stochastic Control Theory 626: 569: 506: 451: 394: 932: 1034:. Vol. 1. IEEE. pp. 855–860. 1091: 983:IEEE Transactions on Automatic Control 314:We can design an observer of the form 974: 837: 86:Consider a deterministic LTI system: 906:M.H.A. Davis and P. Varaiya (1972). 573:{\displaystyle {\dot {e}}=(A-LC)e\,} 458:{\displaystyle u(t)=-K{\hat {x}}\,.} 1026:Atassi, A.N.; Khalil, H.K. (1998). 13: 14: 1115: 513:{\displaystyle e=x-{\hat {x}}\,.} 279:represents the output signal, and 69:linear-quadratic-Gaussian control 848:. Vol. 58. Academic Press. 633:{\displaystyle u(t)=-K(x-e)\,.} 1064: 1019: 623: 611: 599: 593: 563: 548: 500: 445: 427: 421: 370: 361: 346: 332: 297: 291: 266: 260: 235: 229: 196: 190: 174: 168: 158: 152: 140: 134: 118: 112: 1: 970:. Cambridge University Press. 830: 925:10.1016/0022-247X(72)90281-8 248:represents the input signal, 49:linear time-invariant system 7: 912:J. Math. Anal. Applications 24:, more formally known as a 10: 1120: 844:Karl Johan Astrom (1970). 65:linear-quadratic regulator 1005:10.1109/TAC.2013.2259207 1040:10.1109/cdc.1998.760800 941:SIAM Journal on Control 815:together with those of 966:A. Bensoussan (1992). 790: 634: 574: 514: 459: 399: 304: 273: 242: 207: 791: 635: 575: 515: 460: 400: 305: 274: 243: 208: 650: 587: 530: 479: 415: 321: 303:{\displaystyle x(t)} 285: 272:{\displaystyle y(t)} 254: 241:{\displaystyle u(t)} 223: 93: 22:separation principle 1081:Brezinski, Claude. 819: −  811: −  408:and state feedback 67:. This is known as 47:are designed for a 43:and a stable state 1104:Stochastic control 807:are just those of 786: 777: 751: 691: 630: 570: 510: 455: 395: 300: 269: 238: 203: 201: 1085:. Springer, 2002. 989:(10): 2481–2494. 801:triangular matrix 686: 670: 542: 503: 468:Define the error 448: 373: 340: 335: 109: 1111: 1074: 1068: 1062: 1061: 1023: 1017: 1016: 998: 978: 972: 971: 963: 957: 956: 936: 930: 929: 927: 903: 897: 896: 894: 866: 860: 859: 841: 799:Since this is a 795: 793: 792: 787: 782: 781: 756: 755: 696: 695: 688: 687: 679: 672: 671: 663: 639: 637: 636: 631: 579: 577: 576: 571: 544: 543: 535: 519: 517: 516: 511: 505: 504: 496: 464: 462: 461: 456: 450: 449: 441: 404: 402: 401: 396: 375: 374: 366: 342: 341: 336: 328: 326: 309: 307: 306: 301: 278: 276: 275: 270: 247: 245: 244: 239: 212: 210: 209: 204: 202: 111: 110: 102: 1119: 1118: 1114: 1113: 1112: 1110: 1109: 1108: 1089: 1088: 1078: 1077: 1069: 1065: 1050: 1024: 1020: 979: 975: 964: 960: 953:10.1137/0311025 937: 933: 904: 900: 883:10.1137/0309026 871:SIAM J. Control 867: 863: 856: 842: 838: 833: 776: 775: 769: 768: 758: 757: 750: 749: 735: 729: 728: 720: 701: 700: 690: 689: 678: 677: 674: 673: 662: 661: 654: 653: 651: 648: 647: 588: 585: 584: 534: 533: 531: 528: 527: 495: 494: 480: 477: 476: 440: 439: 416: 413: 412: 365: 364: 327: 325: 324: 322: 319: 318: 286: 283: 282: 255: 252: 251: 224: 221: 220: 200: 199: 177: 162: 161: 121: 101: 100: 96: 94: 91: 90: 84: 12: 11: 5: 1117: 1107: 1106: 1101: 1099:Control theory 1087: 1086: 1076: 1075: 1063: 1048: 1018: 973: 958: 947:(2): 323–343. 931: 898: 877:(3): 354–371. 861: 854: 835: 834: 832: 829: 797: 796: 785: 780: 774: 771: 770: 767: 764: 763: 761: 754: 748: 745: 742: 739: 736: 734: 731: 730: 727: 724: 721: 719: 716: 713: 710: 707: 706: 704: 699: 694: 685: 682: 676: 675: 669: 666: 660: 659: 657: 641: 640: 629: 625: 622: 619: 616: 613: 610: 607: 604: 601: 598: 595: 592: 581: 580: 568: 565: 562: 559: 556: 553: 550: 547: 541: 538: 521: 520: 509: 502: 499: 493: 490: 487: 484: 466: 465: 454: 447: 444: 438: 435: 432: 429: 426: 423: 420: 406: 405: 393: 390: 387: 384: 381: 378: 372: 369: 363: 360: 357: 354: 351: 348: 345: 339: 334: 331: 312: 311: 299: 296: 293: 290: 280: 268: 265: 262: 259: 249: 237: 234: 231: 228: 214: 213: 198: 195: 192: 189: 186: 183: 180: 178: 176: 173: 170: 167: 164: 163: 160: 157: 154: 151: 148: 145: 142: 139: 136: 133: 130: 127: 124: 122: 120: 117: 114: 108: 105: 99: 98: 83: 80: 18:control theory 9: 6: 4: 3: 2: 1116: 1105: 1102: 1100: 1097: 1096: 1094: 1084: 1080: 1079: 1072: 1067: 1059: 1055: 1051: 1049:0-7803-4394-8 1045: 1041: 1037: 1033: 1029: 1022: 1014: 1010: 1006: 1002: 997: 992: 988: 984: 977: 969: 962: 954: 950: 946: 942: 935: 926: 921: 917: 913: 909: 902: 893: 888: 884: 880: 876: 872: 865: 857: 855:0-486-44531-3 851: 847: 840: 836: 828: 826: 822: 818: 814: 810: 806: 802: 783: 778: 772: 765: 759: 752: 746: 743: 740: 737: 732: 725: 722: 717: 714: 711: 708: 702: 697: 692: 683: 680: 667: 664: 655: 646: 645: 644: 627: 620: 617: 614: 608: 605: 602: 596: 590: 583: 582: 566: 560: 557: 554: 551: 545: 539: 536: 526: 525: 524: 507: 497: 491: 488: 485: 482: 475: 474: 473: 471: 452: 442: 436: 433: 430: 424: 418: 411: 410: 409: 391: 388: 385: 382: 379: 376: 367: 358: 355: 352: 349: 343: 337: 329: 317: 316: 315: 294: 288: 281: 263: 257: 250: 232: 226: 219: 218: 217: 193: 187: 184: 181: 179: 171: 165: 155: 149: 146: 143: 137: 131: 128: 125: 123: 115: 106: 103: 89: 88: 87: 79: 76: 74: 70: 66: 62: 61:Kalman filter 56: 54: 50: 46: 42: 37: 35: 31: 27: 23: 19: 1082: 1066: 1031: 1021: 986: 982: 976: 967: 961: 944: 940: 934: 915: 911: 901: 874: 870: 864: 845: 839: 820: 816: 812: 808: 798: 642: 522: 469: 467: 407: 313: 215: 85: 77: 57: 38: 25: 21: 15: 918:: 384–402. 825:independent 805:eigenvalues 1093:Categories 892:1808/16692 831:References 34:controller 1058:126270534 996:1103.3005 741:− 712:− 684:˙ 668:˙ 618:− 606:− 555:− 540:˙ 501:^ 492:− 446:^ 434:− 371:^ 353:− 338:˙ 333:^ 107:˙ 1013:12623187 45:feedback 41:observer 30:observer 1056:  1046:  1011:  852:  803:, the 216:where 63:and a 53:stable 1054:S2CID 1009:S2CID 991:arXiv 523:Then 1044:ISBN 850:ISBN 20:, a 1036:doi 1001:doi 949:doi 920:doi 887:hdl 879:doi 16:In 1095:: 1052:. 1042:. 1030:. 1007:. 999:. 987:58 985:. 945:11 943:. 916:37 914:. 910:. 885:. 873:. 827:. 821:LC 813:BK 472:: 75:. 1073:. 1060:. 1038:: 1015:. 1003:: 993:: 955:. 951:: 928:. 922:: 895:. 889:: 881:: 875:9 858:. 817:A 809:A 784:. 779:] 773:e 766:x 760:[ 753:] 747:C 744:L 738:A 733:0 726:K 723:B 718:K 715:B 709:A 703:[ 698:= 693:] 681:e 665:x 656:[ 628:. 624:) 621:e 615:x 612:( 609:K 603:= 600:) 597:t 594:( 591:u 567:e 564:) 561:C 558:L 552:A 549:( 546:= 537:e 508:. 498:x 489:x 486:= 483:e 470:e 453:. 443:x 437:K 431:= 428:) 425:t 422:( 419:u 392:y 389:L 386:+ 383:u 380:B 377:+ 368:x 362:) 359:C 356:L 350:A 347:( 344:= 330:x 298:) 295:t 292:( 289:x 267:) 264:t 261:( 258:y 236:) 233:t 230:( 227:u 197:) 194:t 191:( 188:x 185:C 182:= 175:) 172:t 169:( 166:y 159:) 156:t 153:( 150:u 147:B 144:+ 141:) 138:t 135:( 132:x 129:A 126:= 119:) 116:t 113:( 104:x

Index

control theory
observer
controller
observer
feedback
linear time-invariant system
stable
Kalman filter
linear-quadratic regulator
linear-quadratic-Gaussian control
separation principle in stochastic control
triangular matrix
eigenvalues
independent
ISBN
0-486-44531-3
doi
10.1137/0309026
hdl
1808/16692
"Information states for stochastic systems"
doi
10.1016/0022-247X(72)90281-8
doi
10.1137/0311025
arXiv
1103.3005
doi
10.1109/TAC.2013.2259207
S2CID

Text is available under the Creative Commons Attribution-ShareAlike License. Additional terms may apply.