Model Reference Neural Predictive Controller for Induction Motor Drive - PowerPoint PPT Presentation

1 / 13
About This Presentation
Title:

Model Reference Neural Predictive Controller for Induction Motor Drive

Description:

WSEAS CSCC 05, Vouliagmeni, Athens, Greece. July 11-16, 2005. IM. N. N. P. usaref. ?ref. us ref ... WSEAS CSCC 05, Vouliagmeni, Athens, Greece. July 11-16, 2005 ... – PowerPoint PPT presentation

Number of Views:250
Avg rating:3.0/5.0
Slides: 14
Provided by: Ade9158
Category:

less

Transcript and Presenter's Notes

Title: Model Reference Neural Predictive Controller for Induction Motor Drive


1
Model Reference Neural Predictive Controller for
Induction Motor Drive
University of Quebec at Chicoutimi, Canada
  • Mohand Ouhrouche (Mohand_Ouhrouche_at_uqac.ca)
  • Adel Merabet (Adel_Merabet_at_uqac.ca)
  • Rung-Tien Bui (Rung-Tien_Bui_at_uqac.ca)

WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
2
Contents.
  • Introduction.
  • Induction motor modeling.
  • State space model.
  • Input-output model.
  • Reference control model.
  • Neural predictive Control.
  • Neural networks modeling.
  • Induction motor drive control using neural
    predictive control.
  • Simulation results.
  • Conclusion.

WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
3
Introduction.
  • The rotor speed and flux norm are controlled by a
    neural predictive controller (NPC) with reference
    control model.
  • The reference control model is obtained from the
    machine equations.
  • The method is a combination of artificial neural
    networks (ANN) and predictive control (PC)
    technique.
  • The NPC algorithm is based on the use of ANN as a
    nonlinear prediction model of the motor.
  • The Levenberg-Marquardt optimization method is
    used for ANN training in batch mode.
  • The Newton-Raphson method is used in the
    optimization procedure for predictive control.

WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
4
IM modeling State space model.
Induction motor model in the stationary reference
frame (a , ß)
State equation
The functions f(x), g(x) and h(x) are
sufficiently differentiable.
WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
5
IM modeling Input-output model.
Input-output model
WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
6
IM modeling Reference control model.
Machine equations in complex form
Rotor flux
Where
Reference control model
WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
7
Neural predictive control.
The objective of the predictive control
strategy using neural predictors is twofold
  • To estimate the future output of the plant by
    neural predictor.
  • To minimize a cost function based on the error
    between the predicted output, the reference
    trajectory and the future control error increment
    with reference control model.

WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
8
Neural predictive control.
y (kj)
WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
9
Neural networks modeling for IM.
Neural networks model for induction motor
The training of ANN is done off line by using
Levenberg-Marquardt optimization method, the
procedure is based on batch mode.
WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
10
Induction motor drive control using NPC.
The cost function to be minimized is
with the following assumption
Using the Newton-Raphson (NR) method, the cost
function is minimized iteratively to determine
the best control vector.
WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
11
Simulation results for ANN training.
a. Flux response of motor and ANN b. Motor and
ANN flux error
a. Speed response of motor and ANN b. Motor and
ANN speed error
WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
12
Simulation results for NPC.
a. Reference speed and motor speed b. Tracking
error
a. Reference flux and motor flux norm b. Tracking
error
WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
13
Conclusion.
  • Neural predictive control of induction motor
    drive was presented.
  • Reference control model was obtained from machine
    equations.
  • Prediction algorithm for flux and rotor speed was
    obtained using ANN.
  • Newton-Raphson algorithm was used for optimizing
    the predictive criterion.
  • Simulation results show good response of the
    drive during transient and steady state
    operation.
  • The advantage of this method is the ability of
    the ANN predictor to give good performance
    without requiring knowledge of the machine
    parameters.

WSEAS CSCC 05, Vouliagmeni, Athens, Greece July
11-16, 2005
Write a Comment
User Comments (0)
About PowerShow.com