You are on page 1of 13

Speed Sensorless State Estimation for Induction

Motors: A Moving Horizon Approach


Lei Zhou1 , Yebin Wang2
1

c MERL June 28, 2016


Massachusetts Institute of Technology


Mitsubishi Electric Research Laboratories

1 / 13

Motivations
The tracking bandwidth of speed sensorless induction drive is

limited by the state estimation: estimation scheme with fast


convergence is needed;
The tuning of the classic model reference adaptation estimation is

difficult: systemetic estimation scheme is desired.


The moving horizon estimation (MHE) provides a systematic

framework and demonstrates fast estimation convergence, but is


sensitive to model parameter mismatch.
In this work, we study:
The application of MHE for induction motor speed sensorless

estimation;
Dual-stage adaptive MHE that accounts for parametric model

uncertainties.
c MERL June 28, 2016

2 / 13

Moving Horizon Estimation


Full information estimation:
T
1
X
min
(x0 ) +
Lk (wk , vk ),
T 1
x0 ,{wk }k=0

k=0

where
(x0 ) = kx0 x
0 k2P 1 ,
0

Lk (wk , vk ) = kwk k2Q1 + kyk h(xk )k2R1 .


k

Moving horizon estimation and full


information problem[1] .
[1] Rawlings, James B. Moving horizon
estimation. Encyclopedia of Systems and
Control (2015): 799-804.

c MERL June 28, 2016


Moving horizon estimation:


T
1
X
min
ZT N (z)+
Lk (wk , vk ).

T 1
z,{wk }k=T
N

k=T N

ZT N (z): arrival cost, summarizing the


information from k = 0 to T N .

3 / 13

Induction Motor Model


The classical induction motor model in the stationary two-phase reference frame can
be written as:
1
i ds = ids + dr + qr + uds ,

1
Lm
dr + qr + uqs ,
i qs = iqs

dr = Lm ids dr qr ,
qr = Lm iqs + dr qr ,

TL
= (ids qr + dr iqs )
.
J
J
The parameters are


L2
Rs
Rr
Lm
3 Lm
= Ls 1 m
; =
; =
=
; =
,
Ls Lr

Lr
Lr
2 Lr
where (Rs , Ls ) and (Rr , Lr ) are the resistance and inductance of the stator and rotor
respectively, and Lm is the mutual inductance between stator and rotor.

c MERL June 28, 2016


4 / 13

Adaptive MHE: Problem Formulation


Consider a general nonlinear, discrete system:
xk+1 = fk (xk , uk , wk ),
yk = hk (xk ) + vk .
The MHE formulation for state and parameter estimation at time T is
T
1
X
min
ZT N (z, p) +
Lk (wk , vk , p)
1
p,z,{wk }T
k=T N

subject to

k=T N

xk (k; z, {wj }, uk ) Xk , k = T N, ..., T


wk Wk , k = (T N ), ..., (T 1)
vk = yk hk (xk ) Vk , k = T N, ..., T 1
pP

Remark: the conventional (non-adaptive) MHE does not have parameters p in the
decision variables.
c MERL June 28, 2016

5 / 13

MHE for Induction Motor State Estimation


Stage costs Lk (wk , vk )
Lk (wk , vk ) = wkT Q1 wk + vkT R1 vk .
where Q and R are the covariance matrices for process noise and measurement
noise respectively.
Arrival cost ZT N (z)
Define the cost for the initial estimation error as (x0 ) = (x x
0 )T 1
0 ).
0 (x x
The approximated arrival cost can be calculated by
ZT N (z) = (z x
T N )T 1
T N ) + T N
T N (z x
where T N is updated by the one-step predictor equation as
k+1 = GQGT + Ak k ATk Ak k C T (R + C T k C T )1 Ck ATk .

Constraints
Inequality constraints are removed to simplify the computation. The
optimization problem only subject to the system dynamics.
c MERL June 28, 2016

6 / 13

Dual-stage MHE for parameter adaption


In an augmented-state MHE, estimating parameters and states together can lead
to highly nonlinear optimization problems and add difficulties to solving;

In dual-stage adaptive MHE, the state estimation and parameter estimation are
separated into two sequential steps, and significantly simplify the solving
difficulties.
*
{}*+0
*+,- , {} *+,-

*
{}*+0
*+,2 , {} *+,2

{(}**+,-

State MHE
(# )

+0

Parameter MHE
(%)

State estimation:
T
1
X
x
min
Lx
T =ZT Nx (z)+
k (wk ,vk )
T 1
z,{wk }T N
k=T Nx
x

c MERL June 28, 2016


Parameter estimation:
min
p

T
1
X
pT =p
vkT R1 vk
T Np+
k=T Np

7 / 13

Dual-stage MHE with RLS parameter estimation


Define parameter vector as = [, , , 1/]T .
The current equations can be written into the regression form:

ik+1
ik
ds
ds
k+1dt k
iqs iqs
dt

{z
y

 k
ids
=
ikqs
|
}

k
ds
k
qs

k
qs

k
k
ds
k
{z


ukds

ukqs
} 1/
| {z }

The recursive least square (RLS) estimation can be readily applied as


min

RLS
=
T

T
2
1 X
yk| yk .
T
k=1

Remark: for induction motor parameter estimation (linear, unconstrained system


with no dynamics), RLS estimation is equivalent to MHE parameter estimation
with infinite estimation horizon.
c MERL June 28, 2016

8 / 13

MHE Simulation for Induction Motor


' "#$ +

Flux PI

"#$

+
-

Speed
PI

Torque )
PI

"#$

'

'Inverse Park
Transformation )-

, '-

'
)

Park
Transformation

, )-

2 phase
->
3 phase

.
/
0

., .
3 phase ,
/ /
->
2 phase 0 , 0

Induction
motor

Moving
Horizon
Estimation

Block diagram of induction motor field-oriented control.

c MERL June 28, 2016


9 / 13

MHE Simulation for Induction Motor

Motor Speed (rad/s)

20-steps horizon. Samping rate 10 kHz. P0 = I66 103 . Initial values


x = [1, 1, 0, 0, 5, 0]T ; x
= [0, 0, 0, 0, 0, 0.1]T .
150
100
Reference Speed
Plant Speed
Estimated Speed

50
0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Speed Estimation
Error (rad/s)

6
MHE Speed Estimation Error
EKF Speed Estimation Error

4
2
0
-2

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Time (s)

State estimation for induction motor with MHE and EKF.


c MERL June 28, 2016

10 / 13

MHE Simulation for Induction Motor

Motor Speed (rad/s)

20-steps horizon. Samping rate 10 kHz. P0 = I66 103 . Initial values


x = [1, 1, 0, 0, 10, 0]T ; x
= [0, 0, 0, 0, 0, 0.1]T .
A torque disturbance of 1 Nm is added to motor at T = 0.2 s.
150

100
Reference Speed
Plant Speed Q TL = 1

50

Plant Speed Q TL = 10
Plant Speed Q TL = 100

0.05

0.1

0.15

0.2

0.25

0.3

Speed Estimation
Error (rad/s)

20

0.4

TL

Speed Error Q

TL

= 100

0
-10
-20
1.5

Load torque (Nm)

0.35
Speed Error Q TL = 1
Speed Error Q = 10

10

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

1
True load torque
Estimated, QTL = 1

0.5

Estimated, QTL = 10
Estimated, QTL = 100

0
-0.5

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Time (s)

Induction motor speed with MHE under step torque disturbance.


c MERL June 28, 2016

11 / 13

Dual-stage MHE with RLS for parameter estimation

Parameter
Speed Estimation
Motor Speed (rad/s)
Estimation Error (%)
Error (rad/s)

Initial parameters: 0 = 0.7true ; 0 = 0.7true ; 0 = 0.9true ; 0 = 0.9true .


Dynamic horizon length is used for better convergence.
150
100
Reference
Plant speed
Estimated

50
0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

40
20
0
-20

0.4

50

-50

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Time

Dual-stage adaptive MHE with RLS parameter estimation for induction motor system.

Non-adaptive MHE cannot provide converging estimation.


c MERL June 28, 2016

12 / 13

Conclusion
Moving horizon estimation scheme for induction motor state

estimation considering the rotor mechanical dynamics is studied;


Dual-stage adaptive MHE can successfully achieve converging

parameter and state estimation despite the initial model parametric


error.
Future work
Analysis and tuning for the dual-stage adaptive MHE;
Computational cost reduction of MHE for real-time implementation.

c MERL June 28, 2016


13 / 13

You might also like