Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Save to My Library
Look up keyword
Like this
5Activity
0 of .
Results for:
No results containing your search query
P. 1
Water Quality Parameter Estimation in a Distribution System

Water Quality Parameter Estimation in a Distribution System

Ratings: (0)|Views: 535|Likes:
Published by bryesanggalang

More info:

Published by: bryesanggalang on Jan 19, 2010
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

09/27/2010

pdf

text

original

 
Water Research 39 (2005) 4287–4298
Water quality parameter estimation in a distribution systemunder dynamic state
G.R. Munavalli
1
, M.S. Mohan Kumar
Ã
Department of Civil Engineering, Indian Institute of Science, Bangalore 560 012, India
Received 21 August 2003; received in revised form 23 March 2005; accepted 27 July 2005
Abstract
Chlorine maintenance in distribution systems is an issue for water suppliers. The complex pipe geometry indistribution systems, the dynamic flow conditions experienced within them, and the varied nature of chlorine’sreactivity make it difficult to predict chlorine levels throughout a water system. Computer-based mathematical modelsof water quality transport and fate within distribution systems offer a promising tool for predicting chlorine in a cost-effective manner. Nevertheless, the use of water quality models can only be effective and reliable when both hydraulicsand the mechanisms of chlorine dissipation within the water system are properly defined. Bulk water decay can bemeasured experimentally. However, wall reaction rates are more complex to determine and must be deduced from fieldmeasurement by comparison with simulation results. The simulation–optimization model presented in this paperprovides an effective tool to simplify the chlorine decay model calibration process that is often tedious. Theoptimization tool is based on the weighted-least-squares method solved by Gauss–Newton technique. Application of the model onto a real-life system shows that quantity, quality and location of measurement nodes play an importantrole in estimation of parameters.
r
2005 Elsevier Ltd. All rights reserved.
Keywords:
Chlorine decay modeling; Distribution systems; Drinking water; Optimization tool; Wall reaction parameter
1. Introduction
The study of water quality aspects within a municipalwater distribution system is of great significance as itplays an important role in assuring a good quality of water to the consumer. The maintenance of residualchlorine is used as key criteria to assess the potability of water in the chlorine fed distribution systems. The spreadof chlorine within the distribution system can be beststudied by the use of mathematical models due to thecomplexities arising out of varying hydraulic conditionsand non-applicability of universal chlorine reactionkinetics. The spatio-temporal variations in chlorine levelsare established using forward simulation water qualitymodel. The reaction parameters constitute a vitalcomponent of the input data needed for the realisticsimulations using this model. The predicted chlorineconcentrations within a distribution system are governedby reaction parameters (classified as overall, bulk andwall). However, the task of assigning these parameters tothe pipes individually, globally or zoned is critical. It isgenerally assumed that chlorine decay in the bulk waterentering the distribution system can be described by a
ARTICLE IN PRESS
www.elsevier.com/locate/watres0043-1354/$-see front matter
r
2005 Elsevier Ltd. All rights reserved.doi:10.1016/j.watres.2005.07.043
Ã
Corresponding author.
E-mail addresses:
gurumunavalli@yahoo.co.in(G.R. Munavalli), msmk@civil.iisc.ernet.in(M.S. Mohan Kumar).
1
Presently, Faculty, Department of Civil Engineering,Walchand College of Engineering, Sangli 416 415, India.
 
first-order kinetic model. But the bulk decay parametercan also be non first-order and usually determinedexperimentally by the bottle test. The overall and wallreaction parameters are difficult to measure in the field.Hence the best way of estimating these parameters isthrough calibration against the field measurements.Previously, the trial and error procedure (Clark et al.,1995) was used as a calibration approach. But this is tootedious and may not result in the proper estimation of the parameter values. Also very few studies have beenreported in the literature on methodical estimation of specific parameters.Zeirolf et al. (1998)illustrated theuse of input–output model for chlorine transport toestimate the first-order (global and zoned) wall reactionparameter. The model is applicable only for first-orderreaction kinetics, and does not incorporate storage tanksand multiple water quality sources.Al-Omari andChaudhry (2001)used finite difference procedures forthe determination of overall first-order chlorine decaycoefficient(s).Munavalli and Mohan Kumar (2003)developed an inverse model, which estimates the variousreaction parameters in a multi-source steady-statedistribution system. This model is extended herein forparameter estimation under dynamic state. It essentiallyhas the autocalibration procedure consisting of simula-tion–optimization technique. The Lagrangian forwardsimulation water quality model is utilized in the model.The developed model, which computes the variousreaction parameters in a more direct fashion, is memoryefficient, free from numerical diffusion within the lengthof a segment and avoids the time-consuming trial anderror procedure. In the following sections verification,applicability and usefulness of the inverse model areillustrated using real-life distribution systems.
2. Mathematical model
 2.1. Forward simulation water quality model 
The forward simulation model consists of hydraulicand chlorine transport components which are discussedbelow.
 2.1.1. Hydraulic model 
In the present study, the static hydraulic model (of Niranjan Reddy, 1994) is further modified to computedynamic flows in the pipes using extended periodsimulation.
 2.1.2. Chlorine transport model 
When chlorinated water enters the distribution system,chlorine residual tends to dissipate. Three factors thatfrequently influence chlorine consumption are: (1) reac-tions with organic and inorganic chemicals (e.g., ammo-nia, sulfides, ferrous ion, manganous ion, humic material)in the bulk aqueous phase; (2) reactions with biofilm atthe pipe wall; and (3) consumption by the corrosionprocess (Clark, 1998). Chlorine decay in distributionsystems is generally considered to be composed of twocomponents viz bulk and wall demands. The chlorinetransport model is formulated assuming one-dimensionaladvection-dominated transport phenomenon within apipe segment. Thus, the general governing equation fortransport of chlorine along the
th pipe is given by
@
c
;
t
@
t
¼À
u
@
c
;
t
@
x
À
R
ð
c
;
t
Þ
, (1)where
c
i,t
is the chlorine concentration in pipe
(mg/L) as afunction of distance
x
and time
t
;
u
the mean flow velocityin pipe
(m/s) and
R
(
c
i,t
) the reaction rate expression.
R
(
c
i,t
) represents the combined effect of bulk and wallreactions. The most common combination used is thefirst-order bulk and first-order wall reaction models.However in this paper, the purpose is to provide a largechoice of kinetic models and are discussed in thefollowing paragraphs.(1) Overall first-order reaction kinetics
R
ð
c
;
t
޼0
c
;
t
, (2)where
0
is the overall first-order reaction parameter(d
À
1
).(2) First-order bulk and first-order wall reactions
R
ð
c
;
t
޼b
;
1
c
;
t
À
w
;
1
r
h
ð
w
;
1
þ
Þ
c
;
t
, (3)where,
r
h
is the hydraulic radius (m);
w
,
1
the first-order wall reaction parameter (m/d);
b,1
the first-order bulk decay parameter (d
À
1
) and
the masstransfer coefficient (m/d), the expression for theestimation of this parameter is described in detail byRossman (2000).(3) First-order bulk and zero-order wall reactions
R
ð
c
;
t
޼b
;
1
c
;
t
À
Min
c
;
t
r
h
;
w
;
0
r
h
, (4)where
w,0
is the zero-order wall reaction parameter(mg/m
2
d).(4) Second-order bulk reaction with respect to chlorineonly and first-order wall reaction
R
ð
c
;
t
޼b
;
2
c
2
;
t
À
w
;
1
r
h
ð
w
;
1
þ
Þ
c
;
t
, (5)
b
,
2
is the second-order bulk decay parameter (L/mgd).(5) Second-order bulk reaction with respect to chlorineonly and zero-order wall reaction
R
ð
c
;
t
޼b
;
2
c
;
t
ð
c
;
t
À
L
ÞÀ
w
;
0
r
h
ð
w
;
1
þ
Þ
c
;
t
.(6)
ARTICLE IN PRESS
G.R. Munavalli, M.S. Mohan Kumar / Water Research 39 (2005) 4287–4298
4288
 
(6) Two component second-order bulk decay and first-order wall reactions
R
ð
c
;
t
޼b
;
2
c
;
t
ð
c
;
t
À
L
ÞÀ
w
;
1
r
h
ð
w
;
1
þ
Þ
c
;
t
,(7)where
L
is the limiting concentration of chlorine(mg/L). The value
L
indicates that the chlorinewill decay to a minimum limiting level and nofurther. The non-availability of sufficient organicmatter to react with all the chlorine may be thereason for
L
.(7) Two component second-order bulk and zero-orderwall reactions
R
ð
c
;
t
޼b
;
2
c
;
t
ð
c
;
t
À
L
ÞÀ
Min
w
;
0
r
h
;
r
h
c
;
t
.(8)The mixing of fluid is considered to be completeand instantaneous at the nodes where two or morepipes meet. Water leaving such nodes will carry thechlorine concentration, which is equal to the flow-weighted average of concentrations coming fromthe incoming pipes. Thus the instantaneous andcomplete mixing at the node
and time
t
 j 
is given bythe equationCnc
 j 
;
t
 j 
¼
P
inp
 j 
¼
1
Q
c
;
t
þ
Q
E
E
P
inp
 j 
¼
1
Q
þ
Q
E
;
¼
1
;
. . .
 jn
(9)where
inp
 j 
is the number of incoming pipes at node
j N 
 jn
the total number of nodes in the network,
Q
E
theexternal source flow into at node
(m
3
/s), and
E
theexternal source concentration into at node
(mg/L).The above chlorine transport formulation is solvednumerically by Lagrangian time-driven-method (Liouand Kroon, 1987) and thus the forward simulationmodel is developed. Input parameters needed for thismodel are either an overall first-order reaction para-meter or a bulk reaction parameter along with the wallreaction parameters depending upon the type of reactionrate expressions used.
 2.2. Simulation–optimization mode
As previously mentioned the purpose is to estimatethe global (same for all the pipes in the system) or zoned(multiple) wall reaction parameter(s) from the fieldchlorine measurements. For the zoned parameters, thegrouping can be based on the age, roughness, material,diameter and flow of the pipe. The parameter estimationcan be formulated as an optimization problem where theobjective is to minimize some function of the differencebetween observed and model predicted responses.
 2.2.1. Objective function
The objective function is given byMinimize
¼
X
 j 
¼
1
X
ð
 j 
Þ
¼
1
 j 
;
½
Cno
 j 
;
t
À
Cnc
 j 
;
t
2
, (10)where Cno
 j,tk 
is the observed value of chlorine concen-tration at node
and time
t
(mg/L), Cnc
 j,tk 
thecomputed value of chlorine concentration at node
and time
t
(mg/L),
 j,k 
the weight associated with node
 j 
and time
t
, and
(
 j 
) the number of observationsavailable at node
and
the number of observationnodes.
 2.2.2. Solution technique
The optimization problem is solved by usingGauss–Newton sensitivity analysis technique the detailsof which are given inMunavalli and Mohan Kumar(2003).
 2.2.3. Calibration error statistics
The statistical parameters need to be computed toassess accuracy of the fitting between observed andsimulated concentrations. The parameters computed aretotal number of observations, mean of observedvalues, mean of computed values, mean error, rootmean square (RMS) error, and correlation betweenmeans, for each measurement location and fornetwork as a whole. As the simulation time in thechlorine transport model is advanced by a waterquality time step, it happens that the observation timemay not coincide with the simulation time. Hence it isessential to interpolate the computed chlorine concen-tration corresponding to the observation time. In thepresent study, the required concentration is linearlyinterpolated using the concentrations of end simulationtimes.
 2.2.4. Estimation of parameter uncertainty
Information concerning the uncertainty in parameterestimates is contained in the posteriori parametercovariance matrix (Mishra and Parker, 1989). Thecovariance matrix of estimated parameters also providesinformation regarding the reliability of each of theseparameters. A well-estimated parameter is generallycharacterized by a small variance as compared to aninsensitive parameter that is associated with a largevariance. The more sensitive the parameter, thecloser and quicker the parameter will converge. Acorrelation analysis of the estimated parameters wouldindicate the degree of interdependence among theparameters with respect to the objective function. Afirst-order approximation of the covariance matrix isgiven byBard (1974)
%
s
2
ð
J
T
WJ
Þ
À
1
. (11)
ARTICLE IN PRESS
G.R. Munavalli, M.S. Mohan Kumar / Water Research 39 (2005) 4287–4298
4289

Activity (5)

You've already reviewed this. Edit your review.
1 hundred reads
pankajsemwal86 liked this
pisas liked this
walidabbas liked this
sofyanadnan liked this

You're Reading a Free Preview

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