Sei sulla pagina 1di 5

Summary of TTK4115

Morten Fyhn Amundsen


NTNU

November 29, 2016

1 Matrix stuff
Nullity nul(A) = No. of columns of A rank(A)

Positive definite A symmetric n n real matrix M is positive definite if all its eigen-
values are positive. (Or if zT Mz > 0 for every non-zero vector z of n real numbers.) It
is positive semidefinite if all eigenvalues are positive or zero.

Singularity A square matrix is singular if it is not invertible, i.e. if its determinant


is 0.

Matrix exponential (diagonal)


a 0 e a1 t 0
.1

A = .. .. .. = e At = ...

.. ..
. . . .
a t
0 an 0 e
n

Matrix exponential (Cayley-Hamilton Method)


n1 n1
k Ak with 0 n1 determined by e i t = k ki
X X
e At =
k =0 k =0

Matrix exponential (Laplace method)


( )
e At = L (sI A) 1

Matrix exponential (Jordan form)



e At = Qe At Q 1 where A = Q 1AQ

A2 B . . . An1 B
f g
Controllability matrix C= B AB

gT
CA CA2 . . . CAn1
f
Observability matrix O= C

1
Minimal realisation Given a transfer function; a state-space model that is con-
trollable and observable, and has the same input-output behaviour as the function, is
minimal.

2 Eigen stuff
Eigenvalues Values of such that () = |I A| = 0.

Eigenvectors Vectors v such that (A I)v = 0.

3 Stability
Asymptotic stability Occurs if all poles have strictly negative real parts.

Instability Occurs if one or more poles have positive real parts.

Marginal stability Occurs when the real part of every pole is non-positive, at least
one pole has zero real value, and there are no repeated poles on the imaginary axis.

BIBO stability If bounded input bounded output. Defined for the zero-state
response (initially relaxed system). See Section 4.

Lyapunov stability If every finite initial state gives a finite response. I.e. the zero-
input response.

4 BIBO Stability
4.1 BIBO Stability for Continuous Systems
A continuous system is BIBO stable iff :

(SISO) (t ) is absolutely integrable in [0, ) or |(t )| dt M < for
R
0
some constant M.
has a nega-
(SISO/MIMO) Every pole of every transfer function in G (s) or (s)
tive real part.

4.2 BIBO Stability for Discrete Systems


A discrete system is BIBO stable iff:
has magnitude less than 1.
Every pole of every transfer function in G (s) or (s)

4.3 Lyapunov Stability for Linear Systems


An LTI system x = Ax is stable if there exists a symmetric positive definite matrix P
that satisfies the Lyapunov Equation
AT P + MP = N
Where N is an arbitrary positive definite matrix.

2
5 Discretisation
Z T
Ad = e AT , Bd = e A d B, Cd = C, Dd = D
0

6 Similarity transform
A linear change of coordinates where the original object is expressed with respect to
a different basis. The representation of x with respect to the basis {q1 , . . . , qn } is x
and with Q = [q1 , . . . , qn ], the similarity transform is

x = Q x .

The system originally expressed as

x = Ax + Bu

y = Cx + Du
is transformed to
x = A
x + Bu

y = C x + Du

where
A = Q 1AQ, B = Q 1 B, C = CQ, D = D

7 Jordan canonical form


General strategy:

1. Find all eigenvectors corresponding to an eigenvalue of A.


2. The number of L.I. eigenvectors is the number of Jordan blocks.
3. For each eigenvector q, solve (I A)v = q for the vector v.

8 Statistics

E[X ] = xf (x ) dt
R
Expected value

Variance Var(X ) = E[(X E[X ]) 2 ] = E[X 2 ] (E[X ]) 2

Autocorrelation R X (t 1 , t 2 ) = E[X (t 1 )X (t 2 )]

Wide-sense stationary process X (t ) is WSS if its mean and autocorrelation func-


tions are time invariant: E[X (t )] = and R X (t 1 , t 2 ) = f (t 2 t 1 ).

Spectral density function S X ( ) = F R X ( )




3
GaussMarkov process A stationary Gaussian process X (t ) that has an exponen-
tial autocorrelation is called a GaussMarkov process.

R X ( ) = 2e | |

2 2 2 2
S X ( ) = or S X (s) =
2 + 2 2 s2

9 Linear-quadratic regulator
A system is given as
x = Ax + Bu
with a state feedback u = Kx chosen to minimise the cost function
Z
J= xTQx + uTRu dt
0

where Q is symmetric and positive semidefinite, R is symmetric and positive definite,


and K = R 1 B T P. The matrix P is found by solving

AT P + PA PBR 1 B T P + Q = 0

The relative values of the elements of Q and R enforce tradeoffs between the mag-
nitude of the control action and the speed of the response. The equilibrium can be
shifted from 0 to xeq by instead using u = Pxeq Kx.

10 Kalman filter
The system is given as:
xk +1 = Axk + Buk + wk
zk = Hk xk + vk
The Kalman measurement update equations are:

Kk = PkHkT (Hk PkHkT + Rk ) 1

x k = x k + Kk (zk Hk x k )
Pk = (I Kk Hk )Pk (I Kk Hk ) T + Kk Rk KkT
And the time update equations are:

x k+1 = Axk + Buk

Pk+1 = APk AT + Q k

4
11 Extended Kalman filter
The system is given as:
xk+1 = f (xk , uk ) + wk
yk = h(xk ) + vk
The Kalman measurement update equations are:

Kk = PkCkT (Ck PkCkT + R) 1


 
x k = x k + Kk yk h(x k )

Pk = (I Kk Ck )Pk (I Kk Ck ) T + Kk Rk KkT
And the time update equations are:

x k+1 = f (x k , uk )

Pk+1 = Ak Pk ATk + Q k
Where:
f dh
Ak = and Ck =
xk xk =xk dxk xk =x

k

Potrebbero piacerti anche