Welcome to Scribd. Sign in or start your free trial to enjoy unlimited e-books, audiobooks & documents.Find out more
Download
Standard view
Full view
of .
Look up keyword
Like this
0Activity
0 of .
Results for:
No results containing your search query
P. 1
Modeling and Control of CSTR using Model based Neural Network Predictive Control

Modeling and Control of CSTR using Model based Neural Network Predictive Control

Ratings: (0)|Views: 7|Likes:
Published by ijcsis
This paper presents a predictive control strategy based on neural network model of the plant is applied to Continuous Stirred Tank Reactor (CSTR). This system is a highly nonlinear process; therefore, a nonlinear predictive method, e.g., neural network predictive control, can be a better match to govern the system dynamics. In the paper, the NN model and the way in which it can be used to predict the behavior of the CSTR process over a certain prediction horizon are described, and some comments about the optimization procedure are made. Predictive
control algorithm is applied to control the concentration in a continuous stirred tank reactor (CSTR), whose parameters are optimally determined by solving quadratic performance index using the optimization algorithm. An efficient control of the product concentration in CSTR can be achieved only through accurate model. Here an attempt is made to alleviate the modeling difficulties using Artificial Intelligent technique such as Neural Network. Simulation results demonstrate the feasibility and effectiveness of the NNMPC technique.
This paper presents a predictive control strategy based on neural network model of the plant is applied to Continuous Stirred Tank Reactor (CSTR). This system is a highly nonlinear process; therefore, a nonlinear predictive method, e.g., neural network predictive control, can be a better match to govern the system dynamics. In the paper, the NN model and the way in which it can be used to predict the behavior of the CSTR process over a certain prediction horizon are described, and some comments about the optimization procedure are made. Predictive
control algorithm is applied to control the concentration in a continuous stirred tank reactor (CSTR), whose parameters are optimally determined by solving quadratic performance index using the optimization algorithm. An efficient control of the product concentration in CSTR can be achieved only through accurate model. Here an attempt is made to alleviate the modeling difficulties using Artificial Intelligent technique such as Neural Network. Simulation results demonstrate the feasibility and effectiveness of the NNMPC technique.

More info:

Published by: ijcsis on Aug 19, 2012
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

05/13/2014

pdf

text

original

 
 
Modeling and Control of CSTR using Model based Neural Network Predictive Control
Piyush Shrivastava
Assistant Professor,Electrical& Electronics EngineeringDepartment,Takshshila Institute of Engineering & Technology,Jabalpur, Madhya Pradesh, India.
 Abstract
—this paper presents a predictive control strategy basedon neural network model of the plant is applied to ContinuousStirred Tank Reactor (CSTR). This system is a highly nonlinearprocess; therefore, a nonlinear predictive method, e.g., neuralnetwork predictive control, can be a better match to govern thesystem dynamics. In the paper, the NN model and the way inwhich it can be used to predict the behavior of the CSTR processover a certain prediction horizon are described, and somecomments about the optimization procedure are made. Predictivecontrol algorithm is applied to control the concentration in acontinuous stirred tank reactor (CSTR), whose parameters areoptimally determined by solving quadratic performance indexusing the optimization algorithm. An efficient control of theproduct concentration in cstr can be achieved only throughaccurate model. Here an attempt is made to alleviate themodeling difficulties using Artificial Intelligent technique such asNeural Network. Simulation results demonstrate the feasibilityand effectiveness of the NNMPC technique.
 Keywords-Continuous Stirred Tank Reactor; Neural Network based Predictive Control; Nonlinear Auto Regressive witheXogenous signal.
I.
 
I
 NTRODUCTION
One of the main aims in industry is to reduce operatingcosts. This implies improvements in the final product quality,as well as making better use of the energy resources. Advancedcontrol systems are in fact designed to cope with theserequirements. Model based predictive control (MBPC) [1,2] isnow widely used in industry and a large number of implementation algorithms due to its ability to handle difficultcontrol problems which involve multivariable processinteractions, constraints in the system variables, time delays,etc. The most important advantage of the MPC technologycomes from the process model itself, which allows thecontroller to deal with an exact replica of the real processdynamics, implying a much better control quality. Theinclusion of the constraints is the feature that most clearlydistinguishes MPC from other process control techniques,leading to a tighter control and a more reliable controller.Another important characteristic, which contributes to thesuccess of the MPC technique, is that the MPC algorithmsconsider plant behavior over a future horizon in time. Thus, theeffects of both feedforward and feedback disturbances can beanticipated and eliminated, fact which permits the controller todrive the process output more closely to the referencetrajectory. The classical MBPC algorithms use linear models of the process to predict the output of the process over a certainhorizon, and to evaluate a future sequence of control signals inorder to minimize a certain cost function that takes account of the future output prediction errors over a reference trajectory,as well as control efforts. Although industrial processesespecially continuous and batch processes in chemical and petrochemical plants usually contain complex nonlinearities,most of the MPC algorithms are based on a linear model of the process and such predictive control algorithms may not giverise to satisfactory control performance [3, 4]. Linear modelssuch as step response and impulse response models are preferred, because they can be identified in a straightforwardmanner from process test data. In addition, the goal for most of the applications is to maintain the system at a desired steadystate, rather than moving rapidly between different operating points, so a precisely identified linear model is sufficientlyaccurate in the neighborhood of a single operating point. Aslinear models are reliable from this point of view, they will provide most of the benefits with MPC technology. Even so, if the process is highly nonlinear and subject to large frequentdisturbances; a nonlinear model will be necessary to describethe behavior of the process. Also in servo control problemswhere the operating point is frequently changing, a nonlinear model of the plant is indispensable. In situations like the onesmentioned above, the task of obtaining a high-fidelity model ismore difficult to build for nonlinear processes.In recent years, the use of neural networks for nonlinear system identification has proved to be extremely successful [5-9]. The aim of this paper is to develop a nonlinear controltechnique to provide high-quality control in the presence of nonlinearities, as well as a better understanding of the design process when using these emerging technologies, i.e., neuralnetwork control algorithm. The combination of neuralnetworks and model-based predictive control seems to be agood choice to achieve good performance in the control. In this paper, we will use an optimization algorithm to minimize the
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 10, No. 7, July 201238http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
 cost function and obtain the control input. The paper analyses aneural network based nonlinear predictive controller for aContinuous Stirred Tank Reactor (CSTR), which is a highlynonlinear process. The procedure is based on construction of aneural model for the process and the proper use of that in theoptimization process.This paper begins with an introduction about the predictivecontrol and then the description of the nonlinear predictivecontrol and the way in which it is implemented. The neuralmodel and the way in which it can be used to predict the behavior of the CSTR process over a certain prediction horizonare described, and some comments about the optimization procedure are made. Afterwards, the control aims, the steps inthe design of the control system, and some simulation resultsare discussed.II.
 
P
REDICTIVE
C
ONTROL
The predictive controller, in summary, is characterized bycomputing future control actions based on output values predicted by a model, with vast literature and academic andindustrial interest (Clarke, 1987; Garcia et all, 1989; Arnaldo,1998) [4]. This section presents the concepts of predictivecontrol based on NPC, using the usual optimization functionsand control laws, applied to the conventional predictivecontrollers.
 A.
 
Optimization functions
The optimization function, usually represented by the index J,represents the function that the control action tries tominimize. In an intuitive way, the error between the plantoutput and the desired value is the simplest example of anoptimization function, and it is expressed by:
(1)
 
Where:
 y
(
) represent the plant output
 y k ref 
( )represent the desired response
e
(
) represent the estimation error 
is the sample timeOne of the most usual optimization functions is based on thesquare error and it is represented as:
(2)
But the optimization index can take forms of more complexfunctions. For predictive controllers, whose models arecapable to predict
 N 
steps ahead, the simple application of thesquare error approach can present satisfactory results. Thiscase admits that the optimization function is not limited to anonly point, but an entire vector of 
 N 
 predicted errors. It seeksto optimize the whole trajectory of the future control actions ina horizon of 
 N 
steps ahead.(3)More complex optimization functions can consider the controleffort. It is the specific case of GPC (Generalized PredictiveControl), where the optimization index J can be expressed as:
(4)
where:
 y
(
)
- is the output plant estimation at instant =
Δ
u
- is the control action increment.
 N 
1
- is the minimum horizon of prediction.
 N 
-
is the control horizon.
 N 
- is the maximum horizon of prediction.The objective of the control problem is to minimize the indexJ, with respect to the control actions, looking for the pointswhere the first order differential is null.III.
 
 N
EURAL
 N
ETWORK 
P
REDICTIVE
C
ONTROL
 By the knowledge of the identified neural model of thenonlinear plant which is capable of doing multi step ahead predictions, Predictive control algorithm is applied to controlnonlinear process. The idea of predictive control is tominimize cost function,
 J 
at each sampling point:
[ ] [ ]
u21
22 N Nt=Ni=1
ˆ(,())()y (k+i)(1)
 JtUkrkiuki
 ρ 
=++Δ+
 
(5)With respect to the N
u
future controls,
()[().....(1)]
u
Ukukuk
=+
(6)
 
and subject to constraints:
2
()
u
 NiNn
 
(7)
 
Using the predictive control strategy with identified NARX model (NNMPC) it is possible to calculate the optimalcontrol sequence for nonlinear plant. Here, term
r(k+i)
is therequired reference plant output,
ˆy (k+i)
is predicted NNmodel output,
(1)
uki
Δ+
is the control increment,
 N 
1
and
 N 
2
are the minimum and maximum prediction (or cost)horizons,
 Nu
is the control horizon, and
 ρ 
is the control penalty factor[4].The predictive control approach is also termed as areceding horizon strategy, as it solves the above-definedoptimization problem [5] for a finite future, at a current timeand implements the first optimal control input as the currentcontrol input. The vector 
u = [
Δ
u(k),
Δ
u(k+1),…
Δ
u(k + Nu-1)]
 is calculated by minimizing cost function,
 J 
at each sample
 for selected values of the control parameters {N
1
, N
2
, Nu,
ρ
}.
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 10, No. 7, July 201239http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
Th per timnuis r  Th prostr the
(i
θ 
the
(i
μ 
chasizLesea 
ˆ(H
H[ 
whanconFletheim 
Ti
se controlformance.
 N 
1
ie delay, and ber of time-secursively pre
igure 1: NNM
minimizatio blem minimitegies, iteratiminimum.
+1)()(i)
=+
i
θ μ 
current iterate
i)
is the stepracterized byare selectenberg–Marqrch direction a
ii
[U(t)] +I)d
λ 
ii
G[U(t)] =2[U
ϕ 
=
2i
J(t,(t)] =U(=U(t)
⎛ ⎝ 
ere
 B
(i)
specifi 
G[U 
(i)
(t)]
istrol inputs. Ttcher-Goldfar inverse Hessilementing the
e Series Pred 
arameters ds usually set t
2
is set to defineps in the fut
 
dicted.
C principle ap
of criterion,zed iterativee search met
(i)
.d
 (number ‘i’),size. Variothe way in wd. In theuardt (LM) a pplied in LM
ii
= -G[U(t)]
with Gradie
J(t,U(t))U(t)(t)]E(t)+2
%
2()()
(t))|)ˆY(t)E(t)+2U(t)
i
UtUt 
 ρ 
=
 ⎞ ⎠
es the approxthe gradiente most popul-Shanno (BFan is used heNNMPC is s
iction with Ne
fines the pr o a value 1 the the predictiore for which t
lied to CSTR c
 
in NNMPCily. Similar thods are appl(8) wher 
(i)
d
is the seas types of hich search diresent work lgorithm is ialgorithm is:(9) t vector and H
()()
|U(t)(U(t)
i
UtU
U
=
%%
T
U(t)()|U(t)()
UU
% %
 
imation of theof the
 J 
witar formula knGS) algorithme[8]. The proown in Figure
ral Networks
edictive contat is equal ton horizon i.e.e plant respo 
emical process
an optimizatio NN trainiied to determi,
()
i
θ 
specirch direction algorithms exirection and s Newton ba plemented. Tessian matrix
()()
)|
i
UtUt 
=
(1
()()
i
UtUt 
=
 
(1inverse Hessi
 
h respect town as Broydto approxim posed scheme2.rolheheseonngineiesndist,epedheas:0) 1)
 
ianhen-ateof time scontrol plant o
ahead 
the cunetwor As itinformy
t
is ma rapidIt there better the accan eff effectitime h predictis alsoIThe th
 A.
 
Ge
The darelevaninputfrom tsampledata ar  
 B.
 
 Ne
The fewas cmultila
 
The purposries predictiosignals
u
%
anutput series
 y
 prediction[9],rent control sk will implem
1
ˆ(
 y
+
=
is discussedation for thisltivariable. Oly increasingfore puts highhe model maumulated erroctive methodely reduces trizon. The neion will modereferred to as.
 
M
ODELI
ee steps invol
eration of Inp
ta generated tt informationas given to the conventiod for 0.02 sae obtained to tFigur 
ral Network 
ed forward neosen based oyer perceptronof our neur n of the plantpast data
 y
 N 
.The networ i.e. to predictignal
u
and pnt the functio
,)
t
 y
above, y
t
 rediction to bne problem isivergence dudemands onches the actur. A samplingto reduce thhe number oral network tl the plant. Thystem Identif 
G OF
 N
EURAL
C
ONTROL
(Ned in the AN
ut-Output dat 
train the netabout the dye conventional model, th pling instantain the netwo2: Input-Outp
rchitecture
 
 
twork with sin the trials.l network moutput. Giveit is desired tis trained tothe plant outlant output
 y
nas to contapossible.It isthat this methto accumulatccuracy of thl plant the letime as largeerror accumsteps neededained to do oe acquisitionication.
ETWORK 
P
PC)
 
model devel
 
ork should cnamics of thel model of the input andand the requk.ut Sequencemoidal activaith differentdel is to don a series of o predict thedo
one step
ut
 y
t+1
given. The neural(12)in sufficientassumed thatod will causeion of errors.e model. Thes significantas possible isulation as itfor a givene step aheadf this model
DICTIVE
opment areontain all theCSTR. Thee CSTR andoutput wereired sampledtion functionstructures of 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 10, No. 7, July 201240http://sites.google.com/site/ijcsis/ISSN 1947-5500

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->