21-08-2013, 03:29 PM
ANFIS based Distillation Column Control
ANFIS based Distillation.pdf (Size: 326.83 KB / Downloads: 25)
ABSTRACT
This paper presents a control strategy that combines the predictive
controller and neuro-fuzzy controller type of ANFIS. An Adaptive
Network based Fuzzy Interference System architecture extended
to cope with multivariable systems has been used. The neuro-
fuzzy controller and predictive controller are works parallel. This
controller adjusts the output of the predictive controller, in order
to enhance the predicted inputs. The performance of the control
strategy is studied on the control of Distillation Column problem.
The results confirmed the control quality improvement with MPC
and multi-loop PID controller.
INTRODUCTION
The controller always aims to achieve the process variable to the
given set point value. This is the main task of the properly
designed controller. The controller should also provide some
flexibility in case of change in set point and disturbances. Today
there are many methods for designing intelligent controllers, such
as predictive controller, fuzzy control, neural networks and expert
systems. Various combinations of these controllers give a number
of design possibilities.
Artificial Neural Networks (ANNs) and Fuzzy Logic (FL)
have been increasingly in use in many engineering fields since
their introduction as mathematical aids by McCulloch and Pitts,
1943, and Zadeh, 1965, respectively. Being branches of Artificial
Intelligence (AI), both emulate the human way of using past
experiences, adapting itself accordingly and generalizing. While
the former have the capability of learning by means of parallel
connected units, called neurons, which process inputs in
accordance with their adaptable weights usually in a recursive
manner for approximation; the latter can handle imperfect
information through linguistic variables, which are arguments of
their corresponding membership functions.
MODEL PREDICTIVE CONTROL
Classical MPC
The main idea behind MPC-type controllers is illustrated in
Figure 1 for a SISO system. At sampling time k, a set of m future
manipulated variable moves (control horizon) are selected, so that
the predicted response over a finite horizon p (prediction horizon)
has certain desirable characteristics. This is achieved by
minimizing an objective function based on the deviation of the
future controlled variables from a desired trajectory over the
prediction horizon p and the control energy over the control
horizon m. The MPC optimization is performed for a sequence of
hypothetical future control moves over the control horizon and
only the first move is implemented [8]. The problem is solved
again at time k + 1 with the measured output y (k + 1) as the new
starting point. Model uncertainty and unmeasured process
disturbances are handled by calculating an additive disturbance as
the difference between the process measurement and the model
prediction at the current time step.
Neural Networking Modeling
Neural network needs the system input and output data (Figure 2).
Neural network is connected parallel to the system and they share
input. Second input to neural network is error between system
and neural network output. Based on this error, new parameters
of neural network are adjusted. Sampling period of input and
output data, network architecture, training algorithm and train
periods number are affect the quality of trained neural network.
DISTILLATION COLUMN
The distillation column is probably the most popular and
important process studied in the chemical engineering literature.
Distillation is used in many chemical processes for separating feed
streams and for the purification of final and intermediate product
streams. Most column handle multicomponent feeds, but many
can be approximated by binary or pseudobinary mixtures. The
objective is to split a liquid two component mixture into its
fractions throughout stripping and rectifying processes
DESIGN OF ANFIS CONTROLLER
The basic idea behind the design of neuro-adaptive learning
techniques is very simple. These techniques provide a method for
the fuzzy modeling procedure to learn information about data set,
in order to compute the membership function parameters that best
allow the associated fuzzy inference system to track the given
input-output data. ANFIS constructs an input-output mapping
based on both human knowledge (in the form of fuzzy if-then
rules) and simulated input-output data pairs. It serves as a basis
for building the set of fuzzy if-then rules with appropriate
membership functions to generate the input output pairs.
The parameters associated with the membership functions are
open to change through the learning process. The computation of
these parameters (or their adjustment) is facilitated by a gradient
vector, which provides a measure of how well the ANFIS is
modeling the input-output data for a given parameter set. Once the
gradient vector is obtained, back propagation or hybrid learning
algorithm can be applied in order to adjust the parameters.
Control of the Distillation Column
The control of Distillation column is often considered to be the
benchmark for nonlinear process control because of the highly
nonlinear behavior exhibited by composition dynamics. In this
study, it is aimed to use ANFIS as a controller in a composition
control system. For this purpose, ANFIS controller is designed
and used in an adaptive way in the distillation column control
scheme. Figure 6 illustrates the adaptive control scheme for the
distillation column control system under study. Developed model
is used as a real plant in this scheme. The objective of the system
is to control the composition by manipulating feed flowrate.
Inputs to the controller at each sampling instant are plant and
controller output, Xi(k-1) and F(k-1), respectively at previous
sampling instant. Controller output is the new plant input, F(k).