Research Article Wastewater Treatment Optimization for ...downloads.hindawi.com/journals/mpe/2014/313157.pdf · Wastewater Treatment Optimization for Fish Migration Using Harmony
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Research ArticleWastewater Treatment Optimization forFish Migration Using Harmony Search
Zong Woo Geem1 and Jin-Hong Kim2
1Department of Energy amp Information Technology Gachon University Seongnam 461-701 Republic of Korea2Department of Civil amp Environmental Engineering Chung-Ang University Seoul 156-756 Republic of Korea
Correspondence should be addressed to Jin-Hong Kim jinhongkimcaugmailcom
Received 3 October 2014 Accepted 20 November 2014 Published 4 December 2014
Academic Editor Youqing Wang
Copyright copy 2014 Z W Geem and J-H Kim This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited
Certain types of fish migrate between the sea and fresh water to spawn In order for them to swim without any breathing problemriver should contain enough oxygen If fish is passing along the river in municipal area it needs sufficient dissolved oxygen levelwhich is influenced by dumped amount of wastewater into the river If existing treatment methods such as settling and biologicaloxidation are not enough we have to consider additional treatment methods such as microscreening filtration and nitrificationThis study constructed a wastewater treatment optimization model for migratory fish which considers three costs (filtration costnitrification cost and irrigation cost) and two environmental constraints (minimal dissolved oxygen level and maximal nitrate-nitrogen concentration) Results show that the metaheuristic technique such as harmony search could find good solutions robustlywhile calculus-based technique such as generalized reduced gradient method was trapped in local optima or even divergent
1 Introduction
Various species of fish migrate on time periods ranging fromdaily to annually and over long distances up to thousands ofkilometers [1] Fish normallymigrate because of reproductiveor diet purposes
In order for anadromous fishes such as salmon whichmigrate from the sea into fresh water to spawn to swimupstream without any breathing problem the river shouldcontain enough oxygen or the level of dissolved oxygen (DO)should be more than certain criterion
To control the DO level in a river wastewater dumpedinto the river should be well treated Actually there are severalwastewater treatment techniques such as settling biologicaloxidation microscreening filtration and nitrification [2]
This study intends to find the optimal wastewater treat-ment portfolio which suggests overall minimal cost whilesatisfying minimal DO level over the river reach in order formigratory fishes to swim upstream well
Actually optimal DO control issue for water quality whilelimiting wastewater load has been researched for variousrivers all over the world such asWillamette River in Oregon
USA [3] Schuylkill River in Pennsylvania USA [4] NitraRiver in Slovakia [5] Yasu River in Japan [6] and YamunaRiver in India [7] However real-world problems sometimesrequire model simplification [8] or experience local optimaentrapment [2] This study tries to adopt a more realisticmodel considering nonlinearity of the problem and to findgood results without being entrapped in premature solutions
2 Optimization Formulation
The basic structure of wastewater treatment model in thisstudy came from Haith [2] and can be visualized in Figure 1
A city (population = 100000) dumps wastewater into ariver The wastewater (40000m3day) undergoes secondarytreatment which consists of settling and biological oxidationto remove organic material from the wastewater Howeverthis treatment is not enough to satisfy the water qualitystandard for migratory fishes An environmental regulatoryagency set this standard that the outflow of wastewater treat-ment should be more than 5mgliter (or 5 ppm) of DO insummertime to preserve aquatic life Therefore the city is
Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2014 Article ID 313157 5 pageshttpdxdoiorg1011552014313157
2 Mathematical Problems in Engineering
Boundary40000m3day
Secondary
Filtration
Boundary
Aquifer
River
Percolation
110000m3day
Nitrification
Q2 + Q4Q3 minus Q4
Q4
Q2
Q1
Q3
Q5
Storage
Irrigation A r
Figure 1 Schematic diagram of wastewater treatment portfolio
now forced to consider additional treatment processes thatenhance the water quality
Because the secondary treatment (settling and biologicaloxidation) is not enough the city should consider extra pro-cesses such as filtration and nitrification as shown in Figure 1Also certain amount of wastewater can be diverted to reducethe river discharge This diverted one can be used for cropirrigation
So we can consider total five decision variables for thiswastewater treatment optimization as follows
(1) 1198761 wastewater dumped without additional treatment
Here the cost for filtration 119862ft ($103yr) is represented asfollows
119862ft = 3 (1198762 + 1198764)093
+ 67 (1198762+ 1198764)055
(2)
where the first term in right-hand side of (2) represents thecapital cost of filtration treatment and the second termrepresents operation and maintenance cost
The cost for nitrification 119862nt ($103yr) is represented as
follows
119862nt = 1381198763068
+ 1061198763
042 (3)
where the first term in right-hand side of (3) represents thecapital cost of nitrification treatment and the second termrepresents operation and maintenance cost
The cost for irrigation 119862ir ($103yr) is represented asfollows
119862ir = 219119876028
5+ 12119876
078
5+ 02119876
054
5
+ (131 +48
119903)119876(074+032119903)
5
+ (51 +19
119903)119876(079+028119903)
5minus 068119860
(4)
where the first term in right-hand side of (4) represents thecapital cost of transmission line the second term representsthe capital cost of storage system (lagoon) the third term rep-resents the operation andmaintenance cost of storage systemthe fourth term represents the capital cost of irrigation sys-tem the fifth term represents the operation and maintenancecost of irrigation system and the sixth term represents thenet benefit (negative cost) of cropping (crop sales minus landrent)
Constraints for this optimization problem can be
1198761+ 1198762+ 1198763+ 1198765= 40 (5)
1198763ge 1198764 (6)
0 le 119876119894le 40 119894 = 1 5 (7)
Mathematical Problems in Engineering 3
8 (1 minus 119890minus0063119909
) + 1198620119890minus0063119909
minus 2331198610(119890minus0044119909
minus 119890minus0063119909
)
minus 0671198730(119890minus0025119909
minus 119890minus0063119909
) ge 5
119909 = 5 10 15 20 25 30 40 50
(8)
1198620=8 (110) + 2 (119876
1+ 1198762+ 1198763)
110 + 1198761+ 1198762+ 1198763
(9)
1198610=2 (110) + 25119876
1+ 13119876
2+ 13 (119876
3minus 1198764) + 7119876
4
110 + 1198761+ 1198762+ 1198763
(10)
1198730=5 (110) + 54119876
1+ 50119876
2+ 10 (119876
3minus 1198764) + 10119876
4
110 + 1198761+ 1198762+ 1198763
(11)
654 le 119903 le 1307 (12)
Equations (8) to (11) stand for DO level constraint and (12)stands for nitrate-nitrogen level constraint More detailsabout these constraints are explained in the next section
3 Mathematical Model of Water Quality
Dissolved oxygen 119862(119909) in ppm at a distance (119909 km) down-stream of a point wastewater discharge can be represented asthe following differential equation
119906119889119862
119889119909= 1198962(119862119904minus 119862) minus 119896
1119861 minus 119896119899119873 (13)
where 119906 is river flow velocity (79 kmday in this study) 1198962
is reaeration rate (05day in this study) 119862119904is saturation DO
(80 ppm in this study) 119861 and119873 are remaining carbonaceousbiochemical oxygen demand (CBOD) and nitrogenous bio-chemical oxygen demand (NBOD) (ppm) at distance 119909 and1198961and 119896119899are rate constants (035day and 02day resp) The
first term in the right-hand side of (13) denotes oxygenincrease due to reaeration and the second and third termsdenote oxygen decrease due to oxidation of carbonaceous andnitrogenous material respectively
The above differential equation has the analytic solutionas follows
Equation (14) is identical to (8) where 1198620 1198610 and 119873
0are
respectively river DO CBOD and NBOD right after dis-charge If river water and wastewater are completely mixed atthe discharge point initial 119862
0 1198610 and 119873
0can be calculated
using weighted average as expressed in (9) to (11) where riverflow is 110000m3day river DO is 80 ppm river CBOD is20 ppm and river NBOD is 50 ppm Table 1 shows effluentwater quality after wastewater treatment
Table 1 Effluent water quality after wastewater treatment
Treatment type Effluent quality (ppm)DO CBOD NBOD
Secondary(settling + biological oxidation) 119876
1
2 25 54
Secondary+ filtration (microscreening) 119876
2
2 13 50
Secondary+ nitrification 119876
3minus 1198764
2 13 10
Secondary+ nitrification + filtration 119876
4
2 7 10
For the diverted amount 1198765 we can consider the follow-
ing mass balance equation
Percolation = 119903119879 + 119875 minus ET (15)
where 119879 is irrigation duration (13 weeks in this study) 119875 isprecipitation (cm) during the irrigation season and ET isevapotranspiration (cm) Here 119875 minus ET is zero in this study
A nitrogen balance is used for estimating the nitrogen lossin percolation into groundwater If the nitrogen concentra-tion of119876
5is 119899 (20 ppm in this study) total nitrogen amount to
the irrigation area is 01119903119879119899 (kgha) where 01 is unit conver-sion factor If NC is crop nitrogen uptake (170 kgha in thisstudy) the unused nitrogen of119876
5becomes 01119903119879119899minusNC and
the nitrate-nitrogen concentration 119888119899(ppm) in the percola-
tion becomes as follows
119888119899=nitrogen losspercolation
=119903119879119899 minus 10NC119903119879 + 119875 minus ET
(16)
In this study 119888119899should be less than or equal to 10 ppm for
public health purpose which makes the following constraint
119903119879119899 minus 10NC119903119879 + 119875 minus ET
le 10 (17)
Also the nitrogen contained in1198765should be enough to satisfy
the croprsquos nitrogen requirement as follows
119903119879119899 ge 10NC (18)
From (17) and (18) we can obtain (12) for nitrate-nitrogenconstraint
4 Optimization Using Harmony Search
Thewastewater treatment optimizationmodel for fishmigra-tion constructed in previous sections has a complex structurewhich is difficult to devise a proper search strategy [2] Sothe problem was tackled by commercial software namedMicrosoft Excel Solver which uses the generalized reducedgradient (GRG2) technique [9]
When Solver was applied to the model with five initialsolution vectors it was trapped in local optima or even diver-gent instead of convergent Table 2 shows details
When GRG2 started with the first solution vector (0 0 00 0 0) it was trapped in one of local optima (3349) with
4 Mathematical Problems in Engineering
Table 2 Results by generalized reduced gradient technique
the final solution vector (0 0 40 3033 0 654) when GRG2started using the second solution vector (0 0 0 0 40 10)with the cost of 3458 it did not move any further from itsstarting point when GRG2 started using the third solutionvector (0 0 40 25 0 10) it was trapped in one of local optima(3349) with the final solution vector (0 0 40 3033 0 10)which is identical to the final solution vector of the first case interms of cost but the value of 119903 is different Because119876
5is zero
different 119903 values do not affect the objective function valuewhen GRG2 started using the fourth solution vector (8 0 00 32 10) it was even divergent and disabled to compute anyfurther Although the fourth vector has a good cost (3038)it also slightly violated minimal DO condition at the distanceof 20 km downstream and whenGRG2 started using the fifthsolution vector (10 10 10 5 10 10) with the cost of 3149 it wasalso divergent
On top of those five trials additional trials with differentstarting vectors have been tried But mostly it reached diver-gence instead of convergence because the model structure iscomplicated Thus a metaheuristic approach instead ofcalculus-based approach using the harmony search algo-rithm was introduced to this study
The harmony search (HS) algorithm was inspired bymusic improvisation [10] and applied to various optimizationproblems [11] It has its own unique human-experience-basedderivative [12] For this DO control problem the followingoptimization formulation for HS was used
119862 (119909 1198761 1198762 1198763 1198764) ge 5 ppm
119909 = 5 10 15 20 25 30 40 50(20)
1198761+ 1198762+ 1198763+ 1198765= 40 (21)
1198763ge 1198764 (22)
0 le 119876119894le 40 119894 = 1 5 (23)
654 le 119903 le 1307 (24)
For the above optimization model HS found the optimalsolution using the following steps
Step 1 HS constructs initial memory place named harmonymemory (HM) as in (25) and fills HM with initial solutionvectors as many as harmony memory size (HMS this is 10 inthis study) The initial vectors should satisfy the problemconstraints in (20) to (24)
HM
=
[[[[[
[
1198761
11198761
21198761
31198761
41198761
51199031
119891 (x1)1198762
11198762
21198762
31198762
41198762
51199032
119891 (x2)
119876HMS1
119876HMS2
119876HMS3
119876HMS4
119876HMS5
119903HMS
119891 (xHMS)
]]]]]
]
(25)
Step 2 A new harmony is generated using the followingequation
119909New119894
larr997888
119909119894isin [119909
Lower119894
119909Upper119894
] wp (1 minusHMCR)
119909119894isin HM
= 1199091
119894 1199092
119894 119909
HMS119894
wp HMCR sdot (1 minus PAR)
119909119894+ Δ 119909
119894isin HM wp HMCR sdot PAR
119894 = 1 6
(26)
where HMCR is harmony memory considering rate (095in this study) and PAR is pitch adjustment rate (03 in thisstudy)
Step 3 If the generated vector xNew is better than the worstone xWorst in HM in terms of objective function value thelatter is replaced with the former as follows
xNew isin HM and xWorstnotin HM (27)
Step 4 If termination criterion is satisfied the computation isended Otherwise Step 2 is performed again with updatedHM
When the HS approach was applied to the DO controloptimization model it could successfully find good resultswithout any divergence When ten different runs were per-formed HS found solutions ranging from 3030 to 3117 with
the average of 3075 Table 3 shows more details about thecomputation results
Furthermore this study tackled the above optimizationmodel using genetic algorithm (GA) which is another pop-ular metaheuristic algorithm When GA was applied to thismodel it could find solutions without any divergence Whenten different runs were performed GA found solutions rang-ing from 3033 to 3413 with the average of 3357 AlthoughGA found good solution (3033) only once mostly it foundpremature solutions
5 Conclusions
Thewastewater treatment optimizationmodel for fishmigra-tion was constructed and solved using HS The optimizationmodel considered three costs such as filtration cost nitri-fication cost and irrigation cost and two environmentalconstraints such as minimal DO requirement over riverreach and maximal nitrate-nitrogen concentration for publichealth
While the existing mathematical approach such as GRG2had hard time to identify solutions HS could find bettersolutions without divergence Also HS did not require initialsolution vectors which are very sensitive to final solutionquality When compared with GA HS could find bettersolutions in terms of minimal and average costs
For future study more realistic problems in wastewatertreatment field are expected to be considered and moreupdated techniques for these problems are expected to bedeveloped
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This research was supported by a grant (12-TI-C02) fromAdvanced Water Management Research Program funded byMinistry of Land Infrastructure and Transport of Koreangovernment
References
[1] httpenwikipediaorgwikiFish migration[2] D A Haith Environmental Systems Optimization Wiley New
York NY USA 1982[3] D H Burn and B J Lence ldquoComparison of optimization for-
mulations for waste-load allocationsrdquo Journal of EnvironmentalEngineering vol 118 no 4 pp 597ndash613 1992
[4] H Cardwell and H Ellis ldquoStochastic dynamic program-ming models for water quality managementrdquo Water ResourcesResearch vol 29 no 4 pp 803ndash813 1993
[5] L Somlyody M Kularathna and I Masliev ldquoDevelopment ofleast-cost water quality control policies for theNitra River Basinin SlovakiardquoWater Science andTechnology vol 30 no 5 pp 69ndash78 1994
[6] T Kawachi and SMaeda ldquoDiagnostic appraisal of water qualityand pollution control realities in Yasu River using GIS-aidedepsilon robust optimization modelrdquo Proceedings of the JapanAcademy Series B Physical and Biological Sciences vol 80 no8 pp 399ndash405 2004
[7] A P Singh S K Ghosh and P Sharma ldquoWater quality manage-ment of a stretch of river Yamuna an interactive fuzzy multi-objective approachrdquo Water Resources Management vol 21 no2 pp 515ndash532 2007
[8] D P Loucks C S ReVelle and W R Lynn ldquoLinear program-ming for water pollution controlrdquoManagement Science vol 14no 4 pp B166ndashB181 1967
Figure 1 Schematic diagram of wastewater treatment portfolio
now forced to consider additional treatment processes thatenhance the water quality
Because the secondary treatment (settling and biologicaloxidation) is not enough the city should consider extra pro-cesses such as filtration and nitrification as shown in Figure 1Also certain amount of wastewater can be diverted to reducethe river discharge This diverted one can be used for cropirrigation
So we can consider total five decision variables for thiswastewater treatment optimization as follows
(1) 1198761 wastewater dumped without additional treatment
Here the cost for filtration 119862ft ($103yr) is represented asfollows
119862ft = 3 (1198762 + 1198764)093
+ 67 (1198762+ 1198764)055
(2)
where the first term in right-hand side of (2) represents thecapital cost of filtration treatment and the second termrepresents operation and maintenance cost
The cost for nitrification 119862nt ($103yr) is represented as
follows
119862nt = 1381198763068
+ 1061198763
042 (3)
where the first term in right-hand side of (3) represents thecapital cost of nitrification treatment and the second termrepresents operation and maintenance cost
The cost for irrigation 119862ir ($103yr) is represented asfollows
119862ir = 219119876028
5+ 12119876
078
5+ 02119876
054
5
+ (131 +48
119903)119876(074+032119903)
5
+ (51 +19
119903)119876(079+028119903)
5minus 068119860
(4)
where the first term in right-hand side of (4) represents thecapital cost of transmission line the second term representsthe capital cost of storage system (lagoon) the third term rep-resents the operation andmaintenance cost of storage systemthe fourth term represents the capital cost of irrigation sys-tem the fifth term represents the operation and maintenancecost of irrigation system and the sixth term represents thenet benefit (negative cost) of cropping (crop sales minus landrent)
Constraints for this optimization problem can be
1198761+ 1198762+ 1198763+ 1198765= 40 (5)
1198763ge 1198764 (6)
0 le 119876119894le 40 119894 = 1 5 (7)
Mathematical Problems in Engineering 3
8 (1 minus 119890minus0063119909
) + 1198620119890minus0063119909
minus 2331198610(119890minus0044119909
minus 119890minus0063119909
)
minus 0671198730(119890minus0025119909
minus 119890minus0063119909
) ge 5
119909 = 5 10 15 20 25 30 40 50
(8)
1198620=8 (110) + 2 (119876
1+ 1198762+ 1198763)
110 + 1198761+ 1198762+ 1198763
(9)
1198610=2 (110) + 25119876
1+ 13119876
2+ 13 (119876
3minus 1198764) + 7119876
4
110 + 1198761+ 1198762+ 1198763
(10)
1198730=5 (110) + 54119876
1+ 50119876
2+ 10 (119876
3minus 1198764) + 10119876
4
110 + 1198761+ 1198762+ 1198763
(11)
654 le 119903 le 1307 (12)
Equations (8) to (11) stand for DO level constraint and (12)stands for nitrate-nitrogen level constraint More detailsabout these constraints are explained in the next section
3 Mathematical Model of Water Quality
Dissolved oxygen 119862(119909) in ppm at a distance (119909 km) down-stream of a point wastewater discharge can be represented asthe following differential equation
119906119889119862
119889119909= 1198962(119862119904minus 119862) minus 119896
1119861 minus 119896119899119873 (13)
where 119906 is river flow velocity (79 kmday in this study) 1198962
is reaeration rate (05day in this study) 119862119904is saturation DO
(80 ppm in this study) 119861 and119873 are remaining carbonaceousbiochemical oxygen demand (CBOD) and nitrogenous bio-chemical oxygen demand (NBOD) (ppm) at distance 119909 and1198961and 119896119899are rate constants (035day and 02day resp) The
first term in the right-hand side of (13) denotes oxygenincrease due to reaeration and the second and third termsdenote oxygen decrease due to oxidation of carbonaceous andnitrogenous material respectively
The above differential equation has the analytic solutionas follows
Equation (14) is identical to (8) where 1198620 1198610 and 119873
0are
respectively river DO CBOD and NBOD right after dis-charge If river water and wastewater are completely mixed atthe discharge point initial 119862
0 1198610 and 119873
0can be calculated
using weighted average as expressed in (9) to (11) where riverflow is 110000m3day river DO is 80 ppm river CBOD is20 ppm and river NBOD is 50 ppm Table 1 shows effluentwater quality after wastewater treatment
Table 1 Effluent water quality after wastewater treatment
Treatment type Effluent quality (ppm)DO CBOD NBOD
Secondary(settling + biological oxidation) 119876
1
2 25 54
Secondary+ filtration (microscreening) 119876
2
2 13 50
Secondary+ nitrification 119876
3minus 1198764
2 13 10
Secondary+ nitrification + filtration 119876
4
2 7 10
For the diverted amount 1198765 we can consider the follow-
ing mass balance equation
Percolation = 119903119879 + 119875 minus ET (15)
where 119879 is irrigation duration (13 weeks in this study) 119875 isprecipitation (cm) during the irrigation season and ET isevapotranspiration (cm) Here 119875 minus ET is zero in this study
A nitrogen balance is used for estimating the nitrogen lossin percolation into groundwater If the nitrogen concentra-tion of119876
5is 119899 (20 ppm in this study) total nitrogen amount to
the irrigation area is 01119903119879119899 (kgha) where 01 is unit conver-sion factor If NC is crop nitrogen uptake (170 kgha in thisstudy) the unused nitrogen of119876
5becomes 01119903119879119899minusNC and
the nitrate-nitrogen concentration 119888119899(ppm) in the percola-
tion becomes as follows
119888119899=nitrogen losspercolation
=119903119879119899 minus 10NC119903119879 + 119875 minus ET
(16)
In this study 119888119899should be less than or equal to 10 ppm for
public health purpose which makes the following constraint
119903119879119899 minus 10NC119903119879 + 119875 minus ET
le 10 (17)
Also the nitrogen contained in1198765should be enough to satisfy
the croprsquos nitrogen requirement as follows
119903119879119899 ge 10NC (18)
From (17) and (18) we can obtain (12) for nitrate-nitrogenconstraint
4 Optimization Using Harmony Search
Thewastewater treatment optimizationmodel for fishmigra-tion constructed in previous sections has a complex structurewhich is difficult to devise a proper search strategy [2] Sothe problem was tackled by commercial software namedMicrosoft Excel Solver which uses the generalized reducedgradient (GRG2) technique [9]
When Solver was applied to the model with five initialsolution vectors it was trapped in local optima or even diver-gent instead of convergent Table 2 shows details
When GRG2 started with the first solution vector (0 0 00 0 0) it was trapped in one of local optima (3349) with
4 Mathematical Problems in Engineering
Table 2 Results by generalized reduced gradient technique
the final solution vector (0 0 40 3033 0 654) when GRG2started using the second solution vector (0 0 0 0 40 10)with the cost of 3458 it did not move any further from itsstarting point when GRG2 started using the third solutionvector (0 0 40 25 0 10) it was trapped in one of local optima(3349) with the final solution vector (0 0 40 3033 0 10)which is identical to the final solution vector of the first case interms of cost but the value of 119903 is different Because119876
5is zero
different 119903 values do not affect the objective function valuewhen GRG2 started using the fourth solution vector (8 0 00 32 10) it was even divergent and disabled to compute anyfurther Although the fourth vector has a good cost (3038)it also slightly violated minimal DO condition at the distanceof 20 km downstream and whenGRG2 started using the fifthsolution vector (10 10 10 5 10 10) with the cost of 3149 it wasalso divergent
On top of those five trials additional trials with differentstarting vectors have been tried But mostly it reached diver-gence instead of convergence because the model structure iscomplicated Thus a metaheuristic approach instead ofcalculus-based approach using the harmony search algo-rithm was introduced to this study
The harmony search (HS) algorithm was inspired bymusic improvisation [10] and applied to various optimizationproblems [11] It has its own unique human-experience-basedderivative [12] For this DO control problem the followingoptimization formulation for HS was used
119862 (119909 1198761 1198762 1198763 1198764) ge 5 ppm
119909 = 5 10 15 20 25 30 40 50(20)
1198761+ 1198762+ 1198763+ 1198765= 40 (21)
1198763ge 1198764 (22)
0 le 119876119894le 40 119894 = 1 5 (23)
654 le 119903 le 1307 (24)
For the above optimization model HS found the optimalsolution using the following steps
Step 1 HS constructs initial memory place named harmonymemory (HM) as in (25) and fills HM with initial solutionvectors as many as harmony memory size (HMS this is 10 inthis study) The initial vectors should satisfy the problemconstraints in (20) to (24)
HM
=
[[[[[
[
1198761
11198761
21198761
31198761
41198761
51199031
119891 (x1)1198762
11198762
21198762
31198762
41198762
51199032
119891 (x2)
119876HMS1
119876HMS2
119876HMS3
119876HMS4
119876HMS5
119903HMS
119891 (xHMS)
]]]]]
]
(25)
Step 2 A new harmony is generated using the followingequation
119909New119894
larr997888
119909119894isin [119909
Lower119894
119909Upper119894
] wp (1 minusHMCR)
119909119894isin HM
= 1199091
119894 1199092
119894 119909
HMS119894
wp HMCR sdot (1 minus PAR)
119909119894+ Δ 119909
119894isin HM wp HMCR sdot PAR
119894 = 1 6
(26)
where HMCR is harmony memory considering rate (095in this study) and PAR is pitch adjustment rate (03 in thisstudy)
Step 3 If the generated vector xNew is better than the worstone xWorst in HM in terms of objective function value thelatter is replaced with the former as follows
xNew isin HM and xWorstnotin HM (27)
Step 4 If termination criterion is satisfied the computation isended Otherwise Step 2 is performed again with updatedHM
When the HS approach was applied to the DO controloptimization model it could successfully find good resultswithout any divergence When ten different runs were per-formed HS found solutions ranging from 3030 to 3117 with
the average of 3075 Table 3 shows more details about thecomputation results
Furthermore this study tackled the above optimizationmodel using genetic algorithm (GA) which is another pop-ular metaheuristic algorithm When GA was applied to thismodel it could find solutions without any divergence Whenten different runs were performed GA found solutions rang-ing from 3033 to 3413 with the average of 3357 AlthoughGA found good solution (3033) only once mostly it foundpremature solutions
5 Conclusions
Thewastewater treatment optimizationmodel for fishmigra-tion was constructed and solved using HS The optimizationmodel considered three costs such as filtration cost nitri-fication cost and irrigation cost and two environmentalconstraints such as minimal DO requirement over riverreach and maximal nitrate-nitrogen concentration for publichealth
While the existing mathematical approach such as GRG2had hard time to identify solutions HS could find bettersolutions without divergence Also HS did not require initialsolution vectors which are very sensitive to final solutionquality When compared with GA HS could find bettersolutions in terms of minimal and average costs
For future study more realistic problems in wastewatertreatment field are expected to be considered and moreupdated techniques for these problems are expected to bedeveloped
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This research was supported by a grant (12-TI-C02) fromAdvanced Water Management Research Program funded byMinistry of Land Infrastructure and Transport of Koreangovernment
References
[1] httpenwikipediaorgwikiFish migration[2] D A Haith Environmental Systems Optimization Wiley New
York NY USA 1982[3] D H Burn and B J Lence ldquoComparison of optimization for-
mulations for waste-load allocationsrdquo Journal of EnvironmentalEngineering vol 118 no 4 pp 597ndash613 1992
[4] H Cardwell and H Ellis ldquoStochastic dynamic program-ming models for water quality managementrdquo Water ResourcesResearch vol 29 no 4 pp 803ndash813 1993
[5] L Somlyody M Kularathna and I Masliev ldquoDevelopment ofleast-cost water quality control policies for theNitra River Basinin SlovakiardquoWater Science andTechnology vol 30 no 5 pp 69ndash78 1994
[6] T Kawachi and SMaeda ldquoDiagnostic appraisal of water qualityand pollution control realities in Yasu River using GIS-aidedepsilon robust optimization modelrdquo Proceedings of the JapanAcademy Series B Physical and Biological Sciences vol 80 no8 pp 399ndash405 2004
[7] A P Singh S K Ghosh and P Sharma ldquoWater quality manage-ment of a stretch of river Yamuna an interactive fuzzy multi-objective approachrdquo Water Resources Management vol 21 no2 pp 515ndash532 2007
[8] D P Loucks C S ReVelle and W R Lynn ldquoLinear program-ming for water pollution controlrdquoManagement Science vol 14no 4 pp B166ndashB181 1967
Equations (8) to (11) stand for DO level constraint and (12)stands for nitrate-nitrogen level constraint More detailsabout these constraints are explained in the next section
3 Mathematical Model of Water Quality
Dissolved oxygen 119862(119909) in ppm at a distance (119909 km) down-stream of a point wastewater discharge can be represented asthe following differential equation
119906119889119862
119889119909= 1198962(119862119904minus 119862) minus 119896
1119861 minus 119896119899119873 (13)
where 119906 is river flow velocity (79 kmday in this study) 1198962
is reaeration rate (05day in this study) 119862119904is saturation DO
(80 ppm in this study) 119861 and119873 are remaining carbonaceousbiochemical oxygen demand (CBOD) and nitrogenous bio-chemical oxygen demand (NBOD) (ppm) at distance 119909 and1198961and 119896119899are rate constants (035day and 02day resp) The
first term in the right-hand side of (13) denotes oxygenincrease due to reaeration and the second and third termsdenote oxygen decrease due to oxidation of carbonaceous andnitrogenous material respectively
The above differential equation has the analytic solutionas follows
Equation (14) is identical to (8) where 1198620 1198610 and 119873
0are
respectively river DO CBOD and NBOD right after dis-charge If river water and wastewater are completely mixed atthe discharge point initial 119862
0 1198610 and 119873
0can be calculated
using weighted average as expressed in (9) to (11) where riverflow is 110000m3day river DO is 80 ppm river CBOD is20 ppm and river NBOD is 50 ppm Table 1 shows effluentwater quality after wastewater treatment
Table 1 Effluent water quality after wastewater treatment
Treatment type Effluent quality (ppm)DO CBOD NBOD
Secondary(settling + biological oxidation) 119876
1
2 25 54
Secondary+ filtration (microscreening) 119876
2
2 13 50
Secondary+ nitrification 119876
3minus 1198764
2 13 10
Secondary+ nitrification + filtration 119876
4
2 7 10
For the diverted amount 1198765 we can consider the follow-
ing mass balance equation
Percolation = 119903119879 + 119875 minus ET (15)
where 119879 is irrigation duration (13 weeks in this study) 119875 isprecipitation (cm) during the irrigation season and ET isevapotranspiration (cm) Here 119875 minus ET is zero in this study
A nitrogen balance is used for estimating the nitrogen lossin percolation into groundwater If the nitrogen concentra-tion of119876
5is 119899 (20 ppm in this study) total nitrogen amount to
the irrigation area is 01119903119879119899 (kgha) where 01 is unit conver-sion factor If NC is crop nitrogen uptake (170 kgha in thisstudy) the unused nitrogen of119876
5becomes 01119903119879119899minusNC and
the nitrate-nitrogen concentration 119888119899(ppm) in the percola-
tion becomes as follows
119888119899=nitrogen losspercolation
=119903119879119899 minus 10NC119903119879 + 119875 minus ET
(16)
In this study 119888119899should be less than or equal to 10 ppm for
public health purpose which makes the following constraint
119903119879119899 minus 10NC119903119879 + 119875 minus ET
le 10 (17)
Also the nitrogen contained in1198765should be enough to satisfy
the croprsquos nitrogen requirement as follows
119903119879119899 ge 10NC (18)
From (17) and (18) we can obtain (12) for nitrate-nitrogenconstraint
4 Optimization Using Harmony Search
Thewastewater treatment optimizationmodel for fishmigra-tion constructed in previous sections has a complex structurewhich is difficult to devise a proper search strategy [2] Sothe problem was tackled by commercial software namedMicrosoft Excel Solver which uses the generalized reducedgradient (GRG2) technique [9]
When Solver was applied to the model with five initialsolution vectors it was trapped in local optima or even diver-gent instead of convergent Table 2 shows details
When GRG2 started with the first solution vector (0 0 00 0 0) it was trapped in one of local optima (3349) with
4 Mathematical Problems in Engineering
Table 2 Results by generalized reduced gradient technique
the final solution vector (0 0 40 3033 0 654) when GRG2started using the second solution vector (0 0 0 0 40 10)with the cost of 3458 it did not move any further from itsstarting point when GRG2 started using the third solutionvector (0 0 40 25 0 10) it was trapped in one of local optima(3349) with the final solution vector (0 0 40 3033 0 10)which is identical to the final solution vector of the first case interms of cost but the value of 119903 is different Because119876
5is zero
different 119903 values do not affect the objective function valuewhen GRG2 started using the fourth solution vector (8 0 00 32 10) it was even divergent and disabled to compute anyfurther Although the fourth vector has a good cost (3038)it also slightly violated minimal DO condition at the distanceof 20 km downstream and whenGRG2 started using the fifthsolution vector (10 10 10 5 10 10) with the cost of 3149 it wasalso divergent
On top of those five trials additional trials with differentstarting vectors have been tried But mostly it reached diver-gence instead of convergence because the model structure iscomplicated Thus a metaheuristic approach instead ofcalculus-based approach using the harmony search algo-rithm was introduced to this study
The harmony search (HS) algorithm was inspired bymusic improvisation [10] and applied to various optimizationproblems [11] It has its own unique human-experience-basedderivative [12] For this DO control problem the followingoptimization formulation for HS was used
119862 (119909 1198761 1198762 1198763 1198764) ge 5 ppm
119909 = 5 10 15 20 25 30 40 50(20)
1198761+ 1198762+ 1198763+ 1198765= 40 (21)
1198763ge 1198764 (22)
0 le 119876119894le 40 119894 = 1 5 (23)
654 le 119903 le 1307 (24)
For the above optimization model HS found the optimalsolution using the following steps
Step 1 HS constructs initial memory place named harmonymemory (HM) as in (25) and fills HM with initial solutionvectors as many as harmony memory size (HMS this is 10 inthis study) The initial vectors should satisfy the problemconstraints in (20) to (24)
HM
=
[[[[[
[
1198761
11198761
21198761
31198761
41198761
51199031
119891 (x1)1198762
11198762
21198762
31198762
41198762
51199032
119891 (x2)
119876HMS1
119876HMS2
119876HMS3
119876HMS4
119876HMS5
119903HMS
119891 (xHMS)
]]]]]
]
(25)
Step 2 A new harmony is generated using the followingequation
119909New119894
larr997888
119909119894isin [119909
Lower119894
119909Upper119894
] wp (1 minusHMCR)
119909119894isin HM
= 1199091
119894 1199092
119894 119909
HMS119894
wp HMCR sdot (1 minus PAR)
119909119894+ Δ 119909
119894isin HM wp HMCR sdot PAR
119894 = 1 6
(26)
where HMCR is harmony memory considering rate (095in this study) and PAR is pitch adjustment rate (03 in thisstudy)
Step 3 If the generated vector xNew is better than the worstone xWorst in HM in terms of objective function value thelatter is replaced with the former as follows
xNew isin HM and xWorstnotin HM (27)
Step 4 If termination criterion is satisfied the computation isended Otherwise Step 2 is performed again with updatedHM
When the HS approach was applied to the DO controloptimization model it could successfully find good resultswithout any divergence When ten different runs were per-formed HS found solutions ranging from 3030 to 3117 with
the average of 3075 Table 3 shows more details about thecomputation results
Furthermore this study tackled the above optimizationmodel using genetic algorithm (GA) which is another pop-ular metaheuristic algorithm When GA was applied to thismodel it could find solutions without any divergence Whenten different runs were performed GA found solutions rang-ing from 3033 to 3413 with the average of 3357 AlthoughGA found good solution (3033) only once mostly it foundpremature solutions
5 Conclusions
Thewastewater treatment optimizationmodel for fishmigra-tion was constructed and solved using HS The optimizationmodel considered three costs such as filtration cost nitri-fication cost and irrigation cost and two environmentalconstraints such as minimal DO requirement over riverreach and maximal nitrate-nitrogen concentration for publichealth
While the existing mathematical approach such as GRG2had hard time to identify solutions HS could find bettersolutions without divergence Also HS did not require initialsolution vectors which are very sensitive to final solutionquality When compared with GA HS could find bettersolutions in terms of minimal and average costs
For future study more realistic problems in wastewatertreatment field are expected to be considered and moreupdated techniques for these problems are expected to bedeveloped
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This research was supported by a grant (12-TI-C02) fromAdvanced Water Management Research Program funded byMinistry of Land Infrastructure and Transport of Koreangovernment
References
[1] httpenwikipediaorgwikiFish migration[2] D A Haith Environmental Systems Optimization Wiley New
York NY USA 1982[3] D H Burn and B J Lence ldquoComparison of optimization for-
mulations for waste-load allocationsrdquo Journal of EnvironmentalEngineering vol 118 no 4 pp 597ndash613 1992
[4] H Cardwell and H Ellis ldquoStochastic dynamic program-ming models for water quality managementrdquo Water ResourcesResearch vol 29 no 4 pp 803ndash813 1993
[5] L Somlyody M Kularathna and I Masliev ldquoDevelopment ofleast-cost water quality control policies for theNitra River Basinin SlovakiardquoWater Science andTechnology vol 30 no 5 pp 69ndash78 1994
[6] T Kawachi and SMaeda ldquoDiagnostic appraisal of water qualityand pollution control realities in Yasu River using GIS-aidedepsilon robust optimization modelrdquo Proceedings of the JapanAcademy Series B Physical and Biological Sciences vol 80 no8 pp 399ndash405 2004
[7] A P Singh S K Ghosh and P Sharma ldquoWater quality manage-ment of a stretch of river Yamuna an interactive fuzzy multi-objective approachrdquo Water Resources Management vol 21 no2 pp 515ndash532 2007
[8] D P Loucks C S ReVelle and W R Lynn ldquoLinear program-ming for water pollution controlrdquoManagement Science vol 14no 4 pp B166ndashB181 1967
the final solution vector (0 0 40 3033 0 654) when GRG2started using the second solution vector (0 0 0 0 40 10)with the cost of 3458 it did not move any further from itsstarting point when GRG2 started using the third solutionvector (0 0 40 25 0 10) it was trapped in one of local optima(3349) with the final solution vector (0 0 40 3033 0 10)which is identical to the final solution vector of the first case interms of cost but the value of 119903 is different Because119876
5is zero
different 119903 values do not affect the objective function valuewhen GRG2 started using the fourth solution vector (8 0 00 32 10) it was even divergent and disabled to compute anyfurther Although the fourth vector has a good cost (3038)it also slightly violated minimal DO condition at the distanceof 20 km downstream and whenGRG2 started using the fifthsolution vector (10 10 10 5 10 10) with the cost of 3149 it wasalso divergent
On top of those five trials additional trials with differentstarting vectors have been tried But mostly it reached diver-gence instead of convergence because the model structure iscomplicated Thus a metaheuristic approach instead ofcalculus-based approach using the harmony search algo-rithm was introduced to this study
The harmony search (HS) algorithm was inspired bymusic improvisation [10] and applied to various optimizationproblems [11] It has its own unique human-experience-basedderivative [12] For this DO control problem the followingoptimization formulation for HS was used
119862 (119909 1198761 1198762 1198763 1198764) ge 5 ppm
119909 = 5 10 15 20 25 30 40 50(20)
1198761+ 1198762+ 1198763+ 1198765= 40 (21)
1198763ge 1198764 (22)
0 le 119876119894le 40 119894 = 1 5 (23)
654 le 119903 le 1307 (24)
For the above optimization model HS found the optimalsolution using the following steps
Step 1 HS constructs initial memory place named harmonymemory (HM) as in (25) and fills HM with initial solutionvectors as many as harmony memory size (HMS this is 10 inthis study) The initial vectors should satisfy the problemconstraints in (20) to (24)
HM
=
[[[[[
[
1198761
11198761
21198761
31198761
41198761
51199031
119891 (x1)1198762
11198762
21198762
31198762
41198762
51199032
119891 (x2)
119876HMS1
119876HMS2
119876HMS3
119876HMS4
119876HMS5
119903HMS
119891 (xHMS)
]]]]]
]
(25)
Step 2 A new harmony is generated using the followingequation
119909New119894
larr997888
119909119894isin [119909
Lower119894
119909Upper119894
] wp (1 minusHMCR)
119909119894isin HM
= 1199091
119894 1199092
119894 119909
HMS119894
wp HMCR sdot (1 minus PAR)
119909119894+ Δ 119909
119894isin HM wp HMCR sdot PAR
119894 = 1 6
(26)
where HMCR is harmony memory considering rate (095in this study) and PAR is pitch adjustment rate (03 in thisstudy)
Step 3 If the generated vector xNew is better than the worstone xWorst in HM in terms of objective function value thelatter is replaced with the former as follows
xNew isin HM and xWorstnotin HM (27)
Step 4 If termination criterion is satisfied the computation isended Otherwise Step 2 is performed again with updatedHM
When the HS approach was applied to the DO controloptimization model it could successfully find good resultswithout any divergence When ten different runs were per-formed HS found solutions ranging from 3030 to 3117 with
the average of 3075 Table 3 shows more details about thecomputation results
Furthermore this study tackled the above optimizationmodel using genetic algorithm (GA) which is another pop-ular metaheuristic algorithm When GA was applied to thismodel it could find solutions without any divergence Whenten different runs were performed GA found solutions rang-ing from 3033 to 3413 with the average of 3357 AlthoughGA found good solution (3033) only once mostly it foundpremature solutions
5 Conclusions
Thewastewater treatment optimizationmodel for fishmigra-tion was constructed and solved using HS The optimizationmodel considered three costs such as filtration cost nitri-fication cost and irrigation cost and two environmentalconstraints such as minimal DO requirement over riverreach and maximal nitrate-nitrogen concentration for publichealth
While the existing mathematical approach such as GRG2had hard time to identify solutions HS could find bettersolutions without divergence Also HS did not require initialsolution vectors which are very sensitive to final solutionquality When compared with GA HS could find bettersolutions in terms of minimal and average costs
For future study more realistic problems in wastewatertreatment field are expected to be considered and moreupdated techniques for these problems are expected to bedeveloped
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This research was supported by a grant (12-TI-C02) fromAdvanced Water Management Research Program funded byMinistry of Land Infrastructure and Transport of Koreangovernment
References
[1] httpenwikipediaorgwikiFish migration[2] D A Haith Environmental Systems Optimization Wiley New
York NY USA 1982[3] D H Burn and B J Lence ldquoComparison of optimization for-
mulations for waste-load allocationsrdquo Journal of EnvironmentalEngineering vol 118 no 4 pp 597ndash613 1992
[4] H Cardwell and H Ellis ldquoStochastic dynamic program-ming models for water quality managementrdquo Water ResourcesResearch vol 29 no 4 pp 803ndash813 1993
[5] L Somlyody M Kularathna and I Masliev ldquoDevelopment ofleast-cost water quality control policies for theNitra River Basinin SlovakiardquoWater Science andTechnology vol 30 no 5 pp 69ndash78 1994
[6] T Kawachi and SMaeda ldquoDiagnostic appraisal of water qualityand pollution control realities in Yasu River using GIS-aidedepsilon robust optimization modelrdquo Proceedings of the JapanAcademy Series B Physical and Biological Sciences vol 80 no8 pp 399ndash405 2004
[7] A P Singh S K Ghosh and P Sharma ldquoWater quality manage-ment of a stretch of river Yamuna an interactive fuzzy multi-objective approachrdquo Water Resources Management vol 21 no2 pp 515ndash532 2007
[8] D P Loucks C S ReVelle and W R Lynn ldquoLinear program-ming for water pollution controlrdquoManagement Science vol 14no 4 pp B166ndashB181 1967
the average of 3075 Table 3 shows more details about thecomputation results
Furthermore this study tackled the above optimizationmodel using genetic algorithm (GA) which is another pop-ular metaheuristic algorithm When GA was applied to thismodel it could find solutions without any divergence Whenten different runs were performed GA found solutions rang-ing from 3033 to 3413 with the average of 3357 AlthoughGA found good solution (3033) only once mostly it foundpremature solutions
5 Conclusions
Thewastewater treatment optimizationmodel for fishmigra-tion was constructed and solved using HS The optimizationmodel considered three costs such as filtration cost nitri-fication cost and irrigation cost and two environmentalconstraints such as minimal DO requirement over riverreach and maximal nitrate-nitrogen concentration for publichealth
While the existing mathematical approach such as GRG2had hard time to identify solutions HS could find bettersolutions without divergence Also HS did not require initialsolution vectors which are very sensitive to final solutionquality When compared with GA HS could find bettersolutions in terms of minimal and average costs
For future study more realistic problems in wastewatertreatment field are expected to be considered and moreupdated techniques for these problems are expected to bedeveloped
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgment
This research was supported by a grant (12-TI-C02) fromAdvanced Water Management Research Program funded byMinistry of Land Infrastructure and Transport of Koreangovernment
References
[1] httpenwikipediaorgwikiFish migration[2] D A Haith Environmental Systems Optimization Wiley New
York NY USA 1982[3] D H Burn and B J Lence ldquoComparison of optimization for-
mulations for waste-load allocationsrdquo Journal of EnvironmentalEngineering vol 118 no 4 pp 597ndash613 1992
[4] H Cardwell and H Ellis ldquoStochastic dynamic program-ming models for water quality managementrdquo Water ResourcesResearch vol 29 no 4 pp 803ndash813 1993
[5] L Somlyody M Kularathna and I Masliev ldquoDevelopment ofleast-cost water quality control policies for theNitra River Basinin SlovakiardquoWater Science andTechnology vol 30 no 5 pp 69ndash78 1994
[6] T Kawachi and SMaeda ldquoDiagnostic appraisal of water qualityand pollution control realities in Yasu River using GIS-aidedepsilon robust optimization modelrdquo Proceedings of the JapanAcademy Series B Physical and Biological Sciences vol 80 no8 pp 399ndash405 2004
[7] A P Singh S K Ghosh and P Sharma ldquoWater quality manage-ment of a stretch of river Yamuna an interactive fuzzy multi-objective approachrdquo Water Resources Management vol 21 no2 pp 515ndash532 2007
[8] D P Loucks C S ReVelle and W R Lynn ldquoLinear program-ming for water pollution controlrdquoManagement Science vol 14no 4 pp B166ndashB181 1967