離散時間系統(tǒng)量化動態(tài)輸出反饋的H∞控制

1、 No. 6 CHE Wei-Wei and YANG Guang-Hong: Quantized Dynamic Output Feedback H Control for 657 To facilitate the presentation of Theorem 2, we denote Aeopt = A Bcopt C2 B1 Ccopt , Ceopt = C1 Acopt 0 Acopt 0 Bcopt B1 0 D12 Ccopt By solving LMI (9, a controller is obtained with the gain matrices = Aini c

2、 0.0213 0.9711 1.6320 3.1451 = , B ini c 5.0328 1.3363 0.2317 1opt = B C ini = 0.0040 c Theorem 2. Consider plant (1 controlled by the quantized dynamic output feedback controller (6, if only M1 and M2 are chosen satisfying M1 M2 where opt = opt opt min (Qopt opt 2 C2 min (Qopt (46 (47 2 2 = 1 + ( 3

3、 + Ccopt 1 , opt T + opt min (Qopt with opt = Ceopt D1 + opt + T T T 1 1 Aeopt Popt B1opt and opt = D D1 + 2B opt Popt B1opt . Then, the control strategy (6 with updating 1 and 3 by 2 opt 1 = 2|x c | M1 + opt opt min (Qopt , 3 = 1 (48 correspondingly, the value of is obtained as au = 3.6421. Let = 2

4、 and the quantization errors 1 = 2 = 3 = 0.1. On the one hand, by Lemma 1 and Q( 0 with the ini ini above gain matrices Aini c , B c , C c , and = 4.3346 and = 0.01, we obtain matrices Qlem and Plem . Apparently, min (Qlem = 0.01. By Theorem 1, it is easy to compute /min (Qlem = 279.2494 and 2 C2 /m

5、in (Qlem = 663.0357. Let M1 = 280 279.2494 and M2 = 664 663.0357. According to (7, the range of the quantizer q3 ( can be computed as M3 = 7.5208. In contrast, by Algorithm 1, with Ac0 = Aini c , B c0 = ini B ini c , C c0 = C c , = 0.0001, c0 = 1, c1 = 100, c2 = 10 000, and = 4.3346 and = 0.01, we o

6、btain optimized matrices Qopt , Popt , and the controller gain matrices 0.0179 0.9454 1.4231 2.6540 , B copt = , 1.1833 4.6790 0.2317 . and updating 2 by 2 = 2|y | M2 + opt 2 C2 min (Qopt Acopt = (49 C copt = 0.0040 renders the closed-loop system (8 asymptotically stable and with the H performance b

7、ound . Proof. It is similar to the proof of Theorem 1 and is omitted here. Remark 4. Because min (Q has a signicant eect on the value of /min (Q, the condition Q( 0 is introduced to restrict the value of min (Q, such that min (Q . Remark 5. On the one hand, in Algorithm 1, by means of optimizing the

8、 index , we obtain the optimized solutions Acopt , Bcopt , Ccopt , Popt , and Qopt , such that the values of opt opt /min (Qopt and opt 2 C2 /min (Qopt are minimum. Thus, the ranges M1 , M2 , and M3 can be optimized indirectly. On the other hand, by Theorem 2, when M1 opt opt /min (Qopt and M2 opt 2

9、 C2 /min (Qopt , controller (6 with gain matrices Acopt , Bcopt , and Ccopt , by updating 1 , 3 according to (48 and updating 2 according to (49 renders the closed-loop system (8 asymptotically stable and with the H performance bound . So we have given an optimization method to solve Problem 1. By T

10、heorem 2, we can obtain opt opt /min (Qopt = 244.3255 and opt 2 C2 /min (Qopt = 577.7594. Let M1 = 245 244.3255 and M2 = 578 577.7594. And according to (7, the range of the quantizer q3 ( can be computed as M3 = 6.5810. The quantizer ranges obtained by Theorem 1 and by Theorem 2 will be compared in

11、Table 1. Table 1 Comparison of the quantizer ranges Theorem 1 Theorem 2 245 578 6.5810 M1 M2 M3 280 664 7.5208 3 Example In this section, an example is presented to illustrate the eectiveness of the proposed method. Example 1. Consider the system of form (1 with 0.5 1 B2 = 0 0.5 1 , B1 = , C1 = 1 .5

12、 1 0 0.5 , C2 = 1 0 1 , D 12 = 0 1 1 0 A= From Table 1, it is clear that the ranges of the quantizers q1 (, q2 (, and q3 ( obtained by Theorem 2 are much more improved than the corresponding one obtained by Theorem 1, by reducing 12.58%, 12.19%, and 12.50%, respectively. The initial system state and

13、 controller state are chosen as x 0 = 5, 4 and x c0 = 10, 10, respectively. Let the disturbance input = 10 randn + 5 for k 15, 25, and let the disturbance input = 0 for all the other k, where randn is a normal distribution with mean zero, variance one, and standard deviation one. Then, Fig. 1 shows the regulated output responses of system (8. In Fig. 1, the solid lines show the results obtained by The