• No results found

GMV Control (Generalized Minimum Variance)

N/A
N/A
Protected

Academic year: 2021

Share "GMV Control (Generalized Minimum Variance)"

Copied!
76
0
0

Loading.... (view fulltext now)

Full text

(1)

GMV Control

(Generalized Minimum Variance)

MODEL IDENTIFICATION AND DATA ANALYSIS

Prof. S. Bittanti

POLITECNICO DI MILANO

(2)

GMV

Control

(Generalized Minimum Variance)

n 

Model reference (Q(z) = 0)

n 

Penalized control (P(z) = 1)

(3)

Summary

n 

Example

(minimum phase system)

n  System to be controlled

n  Model reference control design

(4)

Example

n 

System

time domain description ARMAX (1,1,4)

n  y(t) = 0,8y(t – 1) +

+ u(t – 2) + 1,28u(t – 3) + 0,81u(t – 4) + e(t) + 0,6e(t – 1)

(5)

Example

System

n  A(z)y(t) = B(z)u(t-k) + C(z)e(t)

with

n  A(z) = 1 – 0,8z⁻¹

n  B(z) = 1 + 1,28z⁻¹ + 0,81z⁻² k = 2

(6)

System features

n 

Gain:

n  B(1) / A(1) = 15,45

n 

Zeros of A(z) (system poles)

n  z = 0,8

n 

Zeros of B(z):

n  z = 0,64 ± 0.63i

n 

Zeros of C(z):

(7)

Singularities

in the complex plane

n  A(z) x

n  B(z)

(8)

Open loop

step response

(9)

Model reference design

n 

System:

n  A(z)y(t) = B(z)u(t-k) + C(z)e(t)

n  e∼WN(0,s2)

n 

Model reference criterion

n  J = E[ (P(z)y(t+k) - y°(t))²]

(10)

Overall control scheme

H(z) F(z) 1 / G(z) z B(z) 1 / A(z) C(z) yº(t) + u(t) e(t) y(t) - + + C S k

(11)

Model reference design

n 

Controller polynomials

n  F(z)

n  G(z) = PD(z)B(z)E(z)

n  H(z) = C(z)PD(z)

n 

E(z) and F(z) from Long Division

n  PN(z)C(z) = PD(z)A(z)E(z) + z-kF(z)

(12)

Model reference design

n 

Main issue

(13)

Model reference design

n 

Basic idea

P(z) = M(z)

-1

where M(z) is the

reference model

(14)

Reference model

n 

M(z) =

(n poles in – f and gain 1)

n 

Step response of reference model

n 

Time response 90%

(the table shows the number of steps employed by the step response of M(z)to reach 90% of steady state value, as a

(1 + f)

(15)

Reference model choice

f n = 1 n = 2 n = 3 0.1 1 2 2 0.2 2 3 3 0.3 2 3 4 0.4 3 4 5 0.5 4 6 7 0.6 5 8 10 0.7 7 11 14 0.8 11 17 23 0.85 15 28 32 0.9 22 37 50 0.95 45 76 131

(16)

Model reference choice

n  choosing

n  n =2 and f = 0.4

n  ⇒

with such choice 4 steps are required to reach the 90% threshold

(17)

Determination of P(z)

n  Reference model: n  M(z) = n  P(z)= M(z)-1 n  P(z) = (1 – 0.8z-1 + 0.16z-2) / 0.36 n  PN(z) = 1 – 0.8z-1 + 0.16z-2 (1 + 0.4)2 (1 + 0.4z-1)2

(18)

Controller polynomials

n 

2 steps long division leads to

n  E(z) = 2,78 + 1,67z-1 n  F (z) = 0,16 + 0,096z-1 n 

thus:

n  F(z) = 0,16 + 0,096z-1 n  G(z) = PD(z)B(z)E(z) = 1 + 1,88z-1 + 1,58z-2 + 0,49z-3 H(z) = C(z)P (z)

(19)

Overall control scheme

H(z) F(z) 1 / G(z) z B(z) 1 / A(z) C(z) yº(t) + u(t) e(t) y(t) - + + C S k Characteristic Polynomial: B(z)C(z)PN(z)

(20)

Closed loop

step response

(21)

Closed loop

step response

n 

input u(t) with

s

2

= 0

Transfer function from yo to u:

(22)

Penalized control design

n 

System to be controlled:

n  A(z)y(t) = B(z)u(t-k) + C(z)e(t)

n  e∼WN(0,s2)

n 

Performance criterion

n  J = E[(y(t + k) + Q(z)u(t) - yº(t))²]

n  P(z) = 1

(23)

Penalized control design

n 

Controller polynomials

n  F(z) = F˜(z)QD(z)

n  G(z) = B(z)QD(z)E(z) + C(z)QN(z)

n  H(z) = C(z)QD(z)

n 

E(z) and F˜(z) from Long Division

n  C(z) = A(z)E(z) + z-kF˜(z)

(24)

Penalized control design

n 

Closed loop tr function from y° to y

n  S(z) = n 

Characteristic polynomial

n  B(z)QD(z) + A(z)QN(z) z⁻ 1 + Q(z) A(z) B(z) k

(25)

Choice of Q(z)

n 

Typical:

n  Q(z) = q constant n  Q(z) = q (1 - z⁻¹) n  Q(z) = q 1 - z⁻¹ 1 – gz⁻¹

(26)

Choice of Q(z)

n 

Q(z) =

q

n 

Poles: B(z) + q A(z) = 0

n  q = 0 : zeros of B(z)

(27)

Poles

n 

Root locus of B(z) + q A(z)

q = 0

q ≃ 8,57

(28)

Closed loop

step response (q = 1)

n 

Output y(t) with

s

2

= 0

(29)

Closed loop

step response (

q

= 1)

(30)

Closed loop

step response (q = 8,57)

n 

Output y(t) with

s

2

= 0

(31)

Closed loop

step response (q = 8,57)

n 

Input u(t) with

s

2

= 0

(32)

Gain of control system

n 

S(1) =

n 

q

0

gain S(1)

1

non zero

steady state error

1

1 +

q

A(1) B(1)

(33)

Choice of Q(z)

n 

To avoid bias in steady state

Q(z) =

q

(1 - z

¹

)

n 

Poles: B(z) + q(1 - z

¹

)A(z) = 0

n  q = 0 : zeros of B(z)

(34)

Closed loop poles

n 

Root locus of B(z) +

q

(1 - z

¹

)A(z)

q = 0

(35)

Closed loop gain

n 

S(z) =

for z = 1 the gain is 1

n 

Unitary gain guaranteed for all q

1

1 + q(1 - z⁻¹) A(z)

(36)

Closed loop

step response (q = 1)

n 

Output y(t) with

s

2

= 0

(37)

Closed loop

step response (q = 1)

n 

Input u(t) with

s

2

= 0

(38)

Closed loop

step response (

q

= 50)

(39)

Closed loop

step response (q = 50)

n 

Input u(t) with

s

2

= 0

(40)

Choice of Q(z)

n 

Q(z) = q(1 - z

¹

) / (1 – 0.9z

¹

)

n 

Poles:

n  q = 0 : zeros of (1 – 0.9z⁻¹)B(z)

(41)

Closed loop poles

n  Root locus of (1 – 0.9z⁻¹)B(z) + q(1 - z⁻¹)A(z)

q → ∞

(42)

Closed loop

step response (q = 1)

n 

Output y(t) with

s

2

= 0

(43)

Closed loop

step response (q = 1)

n 

Input u(t) with

s

2

= 0

(44)

Closed loop

step response (q = 7,3)

n 

Output y(t) with

s

2

= 0

(45)

Closed loop

step response (q = 7,3)

n 

Input u(t) with

s

2

= 0

(46)

Choice of Q(z)

n 

Q(z) = q(1 - z

¹

) / (1 – 0.8z

¹

)

n 

Closed loop poles:

n  q = 0 : zeros of (1 – 0.8z⁻¹)B(z)

(47)

Closed loop poles

n  Root locus of (1 – 0.8z⁻¹)B(z) + q(1 - z⁻¹)A(z)

q = 0

(48)

Closed loop

step response (q = 6,1)

n 

Output y(t) with

s

2

= 0

(49)

Closed loop

step response (q = 6,1)

n 

Input u(t) with

s

2

= 0

(50)

Closed loop

step response (q = 6,1)

n 

Output y(t) with

s

2

= 10

-4

(51)

Closed loop

step response (q = 6,1)

n 

Input u(t) with

s

2

= 10

-4

(52)

GMV Control

(Generalized Minimum Variance)

MODEL IDENTIFICATION AND DATA ANALYSIS

Prof. S. Bittanti

POLITECNICO DI MILANO

(53)

Ezample 2:

non-minimum phase system

n 

System ARMAX (1,2,3)

n  A(z) = 1 - 0,5z⁻²

n  B(z) = 1 – 2z⁻¹ + 2z⁻² + z⁻³ k = 1

n  C(z) = 1 – 1,4z⁻¹ + 0,7z⁻²

(54)

System features

n  Gain

n  B(1) / A(1) = 4

n  Zeros of A(z) (poles)

n  z = 0,71 n  z = -0,71 n  Zeros of B(z): n  z = -0,35 n  z = 1,18 ± 1,20i n  Zeros of C(z):

(55)

Singularities

in the complex plane

n  A(z) x

n  B(z)

(56)

Open loop

step response

(57)

Choosing Q(z)

n 

With: Q(z) = q(1 - z

¹

) / (1 – 0,5z

¹

)

n 

Poles:

n  q = 0 : zeros of (1 – 0.5z⁻¹)B(z)

(58)

Closed loop poles

n  Root locus of (1 – 0.5z⁻¹)B(z) + q(1 - z⁻¹)A(z)

q = 0

(59)

Closed loop

step response (

q

= 1,5)

(60)

Closed loop

step response (

q

= 1,5)

(61)

Closed loop

step response (

q

= 2,1)

(62)

Closed loop

step response (

q

= 2,1)

(63)

Closed loop

step response (

q

= 19)

(64)

Closed loop

step response (

q

= 19)

(65)

Closed loop

step response (

q

= 19)

(66)

Closed loop

step response (

q

= 19)

(67)

GMV Control

(Generalized Minimum Variance)

MODEL IDENTIFICATION AND DATA ANALYSIS

Prof. S. Bittanti

POLITECNICO DI MILANO

(68)

Example 3

n 

Non-minimum phase system

(zeros outside the unit disk)

and

oscillatory open loop behaviour

(poles located near the border of

stability region)

(69)

Singularities

n  Zeros of A(z) (open loop poles)

n  z = - 0,95 ± 0,1i

n  z = -0,5 ± 0,6i

n  Zeros of B(z) (open loop seros)

n  z = 1 ± i

n  z = 0,2 ± 0,6i

n  Zeros of C(z):

(70)

ARMAX model

n  (4,1,4)

n  A(z) = 1 + 2,9z⁻¹ + 3,422z-2 + 2,072z-3 + 0,557z-4

n  B(z) = 1 – 2,4z-1 + 3,2z-2 – 1,6z-3 + 0,8z-4

n  C(z) = 1 + 0,7z⁻¹

n  Input output delay: k = 1

n  Time domain equation

n  y(t) = -2,9y(t-1) - 3,422y(t-2) - 2,072y(t-3) -

0,557y(t-4) + + u(t-1) - 2,4u(t-2) + 3,2u(t-3) - 1,6u(t-4) + 0,8u(t-5) +

(71)

Singularities

in the complex plane

n  A(z) x

n  B(z)

(72)

Open loop

step response

(73)

Choosing Q(z)

n 

Q(z) =

q

(1 - z

¹

)/(1 + 0,3z

-1

)

n 

Closed loop poles:

n  q = 0 : zeros of (1 + 0,3z-1)B(z)

(74)

Closed loop poles

n  Root locus of (1 + 0,3z-1)B(z) + l(1 - z⁻¹)A(z)

l → ∞

(75)

Closed loop

step response (q = 30)

n 

Output y(t) with

s

2

= 0

(76)

Closed loop

step response (q = 30)

n 

Input u(t) with

s

2

= 0

References

Related documents

Interestingly, all participants who already had the experience of living abroad claimed to have experienced RCT because of learning English, which makes it plausible to suggest that

Genocidal sexual assault on women and the role of culture in the rehabilitation process: Experiences from working with Yazidi women in Turkey.. Sahika Yüksel*, Suzan Saner**,

Vol 12, Issue 8, 2019 Online 2455 3891 Print 0974 2441 THE COMBINED EFFECT OF DEXAMETHASONE AND BACILLUS CALMETTE?GU?RIN ON THE NEUROBEHAVIORAL ASPECT IN MALE MICE (MUS MUSCULUS)

The highest plant height, dry matter accumulation, leaf area index, crop growth rate, seed yield and stick yield were documented with the application of straw mulching @ 7

Quantifying the performance of models trained on labeled data for a single variety (e.g., the majority variety Bokmål) when applied to data from the other (Nynorsk), we found

Measuring the diversity between the location and content ontology based results with similar in search engine .here the PMSE consist of a content feature and a location feature

Methods: Using in-depth interviews with adult Chinese, Malay, and Indian Singaporeans conducted in English/mother-tongue, subjects were shown a health state with 6 levels (Health

Since it is for the Member States to determine the conditions under which legal recognition is given to the change of gender of a person in R.'s situation-as the