Foreleser: Ivan Kalaykov, Universitetet i Tromsų
Sted: Rom B343, Elektro B
Tid: Onsdag 3. Desember, kl 1415-1500
NEURAL NETWORK MODELLING AND CONTROL OF PARAMETER DEPENDING PLANTS
Ivan Kalaykov, Universitetet i Tromsų
This method is appropriate for stochastic plants with fast parameter
changes without possibility for stabilisation of them near stationary point
of the dependent variable. Simulations of a superheater behaviour in power
plant load-frequency control systems with large and fast changes of the
load are presented.
Comparative analysis is carried out of studied NN depending on parametric
disturbance time characteristics, availability to maintain special
operation conditions, learning time, accuracy. Dynamic NN model based
predictive control (NNMBPC) scheme is proposed in case with fast parametric
disturbance deviations. A case study of a NNMBPC of a muffle continuos
patenting furnace is considered.
Processes with dynamical models having parameters depending on internal or
external variables constitute a particular class of plants. As these
variables influence the plant parameters, they are called here paper
parametric disturbances. The mostly used technique to compensate these
disturbances is the gain scheduling approach for adaptive control. It
however generally shows many drawbacks and causes stability problems. The
other possibility to treat the overall system from non-linear system theory
brings up rather complex solutions that cannot cope with real-time
implementations. So, it appears that specific methods of modelling and
simulation for parameter depending plants can give definite advantages than
the general approaches for non-linear plants.
The modelling and identification of parameter depending plants is mostly
based on the assumption the parameters are "frozen" from time to time.
However, this is valid in case of slow variations of the parametric
disturbance towards the nominal plant model. Successful identification in
case of fast plant variations is available for lower order plants in model
reference schemes. The application of artificial NN for modelling and
identification of parameter depending plants is slightly investigated. So
in this work we present an approach of building a NN models with specific
structures that better cope with the plant physical properties. Different
static and dynamic NN structures are proposed and later illustrated by
application examples of steam boiler superheater with the relative steam
flow considered as parametric disturbance, and some metallurgical plants
like patenting furnaces. Comparative analysis is provided to clarify the
relative properties of various NN depending on character of parametric
disturbance variations, possibilities for maintenance special operation
conditions to learn NN, learning time, accuracy. Dynamic NN model based
predictive control (NNMBPC) scheme is presented to overcome the problems
with fast variations of the parametric disturbance which is part of the
normal plant operation.
------------------------------------------------------------------------
To cancel your subscribtion/subscribe to this list: Send mail to
majordomo@chembio.ntnu.no wiht "UN-/SUBSCRIBE prost" in the messagebody.