Excitation and Turbine Adaptive Critic Based Neurocontrol of Multiple Generators on the Electric Power Grid

1

• Excitation and Turbine Adaptive Critic Based
  Neurocontrol of Multiple Generators on the
  Electric Power Grid
• Ganesh Kumar Venayagamoorthy
• Department of Electrical and Computer
  Engineering
• University of Missouri-Rolla, USA
• Acknowledgement
• Ron G Harley Donald C Wunsch
• NSF-USA University of Natal-South Africa

2
Presentation Outline

• Electric Power System its challenges
• Single Machine Power System
• Indirect Adaptive Control Approach
• Adaptive Critics Approach
• Results
• Multimachine Power System
• Adaptive Critics Approach
• Results
• Conclusions

3
Electric Power Grid
Loads
M
Busbars
Lines
Generators
M
4
Multi-Machine Power System

5
Multi-Machine Power System

6
A Single Machine Infinite Bus System
Governor
Generator
Line
Pm
Turbine
VB
VT
Exciter
Vfield
?
Vref
AVR
VAVR
PSS
??
VPSS
7
Conventional Controllers
8
Conventional Controllers

• A turbogenerator - nonlinear, non-stationary,
  fast acting, MIMO device, wide range of operating
  conditions and dynamic characteristics.
• Conventional AVRs, PSSs and turbine governors
• linearized power system model
• one operating point (OP).
• At any other Ops
• performance of the controllers degrades
• undesirable operating states.

9
Improving the Performance of Conventional AVR
and Turbine Governor Controllers

• Neural network based PSSs OP Malik 1994, YM Park
  1996
• Inverse model
• fidelity of the model
• robustness is questionable
• Neural network based PSSs KY Lee 1996
• Indirect adaptive control NN identifier and
  controller
• learning based on one time step error
• online learning required
• Tabu search and genetic algorithms for optimal
  PSS parameters Abido 1999, 2000
• Fuzzy logic PSSs Malik
• Venayagamoorthy Indirect Adaptive Neural
  Network controller
• AVR and governor
• Online training
• Simulation and experimental 1999

10
Adaptive Neurocontroller
11
Adaptive Critic Designs based Controllers

• Adaptive Critic Designs (ACDs) may be defined as
  designs that attempt to approximate dynamic
  programming in the general case.
• Dynamic Programming, in turn, is the only exact
  and efficient method for finding an optimal
  strategy of action over time in a noisy,
  nonlinear, non-stationary environment.
• The basic concept of all forms of dynamic
  programming can be illustrated as follows

12
Dynamic Programming

• Bellmans equation of dynamic programming
• The cost of running true dynamic programming is
  proportional (or worse) to the number of possible
  states in the plant that number, in turn, grows
  exponentially with the number of variables in the
  environment - curse of dimensionality
• Therefore the need for approximate dynamic
  programming methods- ACDS

13
Adaptive Critic Designs

• Adaptive Critic Designs (ACDs) may be defined as
  designs that attempt to approximate dynamic
  programming in the general case.
• Critic Action networks

14
Forms of Adaptive Critic Designs

• Model dependent Adaptive Critics
• Action dependent Adaptive Critics

15
HDP Critic Neural Network
16
HDP based Action Neural Network
17
HDP Neurocontroller Designs

• Action Network Learning

• Critic Network Learning

18
DHP Critic Network Adaptation
DHP critic networks learn minimization of the
following error measure over all time steps, t
19
DHP Critic Network Adaptation
20
Action Network Adaptation

21
DHP Action Network Adaptation
22
A Single Machine Infinite Bus System
Governor
Generator
Line
Pm
Turbine
VB
VT
Exciter
Vfield
?
Vref
AVR
VAVR
23
Rotor Angle for a Temporary Short Circuit
75
70
DHP
HDP
65
CONV
60
Rotor angle (degrees)
55
50
45
40
35
30
0
1
2
3
4
5
6
Time in seconds
24
Rotor Angle for a Temporary Short Circuit
DHP
HDP
80
COT
CONV
75
70
Load angle in degrees
65
60
55
50
45
0
1
2
3
4
5
6
7
Time in seconds
25
Micro-Machine Research Laboratory at the
University of Natal, Durban, South Africa
26
Rotor Angle for a Temporary Short Circuit
27
Multimachine Power System
??1
4
1
7
S2
Governor
3
Micro 1
?Pref1
?
G1
0.01
j0.5
j0.25
Pm1
0.012
j0.75
0.022
Micro-Turbine
Exciter
Pref1
S3
S1
VE1
??1
Vpss

?
PSS
0.022

Load
Vt1
Infinite Bus
Vref1
AVR
j0.75
5
2
Governor
??2
Micro 2
?Pref2
6
?
G2
Pm2
j0.50
0.012
j0.25
Micro-Turbine
0.01
Exciter
Pref2
VE2
Vref2
AVR
Vt2
28
Micro-Machine Research Laboratory at the
University of Natal, Durban, South Africa
29
Micro-Machine Research Laboratory at the
University of Natal, Durban, South Africa
December 2000/ January 2001
30
Block Diagram of the Laboratory System Setup
31
M62/67 DSP Card

• TMS320C6701- 32 bit floating point DSP
• 15 times faster than any other single DSP
• Expansion - two OMNIBUS I/O module cards
• Allows for custom module designs
• Plugs into a standard 32-bit PCI bus slot
• Multiple cards may be installed in systems
  with full driver support under Windows 9x and NT

32
Multimachine Power System
4
7
1
1
Pref1
S2
Micro 1
3
Pm1
?
G1
j0.5
j0.25
0.01
0.012
j0.75
0.022
Micro-Turbine
?Pref1
Exciter
VE1
Ve1
?

?

Ve1
0.022
DHP Neurocontroller
??1
Infinite Bus
j0.75
5
?Vt1
2
Governor
??2
Micro 2
?Pref2
6
?
G2
Pm2
j0.50
0.012
j0.25
Micro-Turbine
0.01
Exciter
Pref2
VE2
Vref2
AVR
Vt2
33
Machine 1 Trans. Line Impedance Increase
1.01
1
Terminal voltage in pu
0.99
CONV_CONV
0.98
CONV_PSS_CONV
0.97
DHP_CONV
10
10.5
11
11.5
12
12.5
13
13.5
14
14.5
Time in seconds
40
35
Load angle in degrees
CON_CONV
30
CONV_PSS_CONV
25
DHP_CONV
20
10
10.5
11
11.5
12
12.5
13
13.5
14
14.5
15
Time in seconds
34
Machine 2 Trans. Line Impedance Increase
40
35
Load angle in degrees
CON_CONV
CONV_PSS_CONV
DHP_CONV
30
10
12
14
16
18
20
22
Time in seconds
35
Multimachine Power System
4
7
1
S2
Pref1
Micro 1
3
Pm1
?
G1
j0.25
0.01
j0.5
0.012
j0.75
0.022
Micro-Turbine
?Pref1
Exciter
S1
VE1
Ve1
?

?

Ve1
Load
0.022
DHP Neurocontroller 1
??1
Infinite Bus
j0.75
5
?Vt1
2
Pref2
Micro 2
Pm2
6
?
G2
j0.50
0.012
j0.25
Micro-Turbine
?Pref2
0.01
Exciter
VE2
?Ve2


Ve2
DHP Neurocontroller 2
??2
?Vt2
36
Machine 2 1 Inductive Load Addition
45
40
35
Load angle in degrees
30
25
10
11
12
13
14
15
16
17
18
19
20
Time in seconds
50
45
40
35
Load angle in degrees
30
25
20
15
10
15
20
25
Time in seconds
37
Machine 2 1 Trans. Line Impedance Increase
40
Load angle in degrees
CONV__CONV
35
CON_PSS_CON
DHP_CONV
DHP_DHP
30
10
12
14
16
18
20
22
Time in seconds
40
35
Load angle in degrees
30
25
20
10
10.5
11
11.5
12
12.5
13
13.5
14
14.5
15
Time in seconds
38
Machine 2 1 Short Circuit
50
45
Load angle in degrees
40
10
11
12
13
14
15
16
17
18
19
20
Time in seconds
50
45
Load angle in degrees
40
35
10
11
12
13
14
15
16
17
Time in seconds
39
Conclusions

• A new method, based on adaptive critics for the
  design of neurocontrollers for generators in a
  multi-machine power system has been implemented
  in simulation and in a laboratory environment.
• All control variables are based on local
  measurements, thus, the control is decentralized.
• The results show that the critic based
  neurocontrollers ensure superior transient
  response throughout the system, for different
  disturbances and different operating conditions,
  compared to a conventional AVR and governor when
  equipped with a PSS.
• The use of such intelligent nonlinear controllers
  will allow power plants on the electric power
  grid to operate closer to their stability limits
  thus producing more power per invested Dollar!

40
What Next?
41
Multi-Machine Power System