S. Mohsen Sadatiyan A.,Samuel Dustin Stanley,Donald V. Chase,Carol J. Miller,Shawn P. McElmurry
Optimizing Pumping System for Sustainable Water Distribution
Networkby Using Genetic Algorithm
Energy & WaterEnergy and water issues are linked together
About 5% of energy demand of US is related to water supply and treatment
About 75% of operation costs of municipal water facilities are attributed to energy
demand
Energy Extraction & generation requires
waterWater Extraction,
treatment & distribution requires
energy
Optimal Pumping Schedule
reduce total pumping cost
shift pump operation time &
space
change in energy cost by
time
optimal pump schedule
minimum energy demand, cost &
associated pollutant emissions
reduce pollutant emission
shift energy demand time &
space
change in pollution
emission by timemeet
system requirement
s with different set of operation schedules
Multi-Objective & Multi-Criteria Optimization
Optimization Methods
Traditional Analytical Methods
Evolutionary Algorithms
fitness of solution
Global Optimumderivatives or other auxiliary characteristics
may results Local Optimum
Genetic Algorithm
•pumping schedule•genetic analogy
• the best solution of the last generation=optimum solution
Fitness evaluatio
n & Elitist
Reproduction
(Crossover)
Mutation
Optimizing Software and Case Studies
PEPSO: Pollutant Emission & Pump Station Optimization
2 drinking water systems within the Great Lakes watershed
PEPSO V4.0~4.5
PEPSO V8.0~8.0.3
Visual interface
Modified Crossover & Mutation
Quasi-Newto
n Metho
dMulti-
Objective
Variable
speed pump
Genetic Algorith
m
DiscreteVs.
Continuous
PEPSO V1.0~3.0
Continuous Method
Discrete Method
Discrete & Continuous Methods
Memory Usage of Continuous Method
Mc= memory usage (byte)n= number of pumpsc= number of cycle per modeling duration2 bytes= required memory for storing a number between 0 to 86400 second (for greater time intervals or shorter modeling period, 1 byte can be used)
Memory Usage of Discrete Method
Md= memory usage (byte)n= number of pumpsT= duration of modelingI= time intervals1 byte/8= 1 bit (“0” or “1” – ON or OFF)
Crossover of Continuous Method
Mutation of Continuous Method
• Mutation
•infeasible children•pairs of controls instead of one control•sorting solution arrays by time•remaining problem for near optimum solutions
Crossover of Discrete Method
• Crossover• multipoint crossover• Identical breaking points for both parents
• Does not have time infeasibility
Mutation of Discrete Method
• Mutation• invert randomly selected gene• replace randomly selected gene by random
number
Variable Speed Pumps
• A random number between min & max speed ratio for mutation
Continuous Method
a column for speed ratio of pump for
each cycle
Discrete Methodreplace OFF=0 and ON=1, by fractional
numbers (speed ratio of pumps)
Existing PEPSO & New Research AreasPEPSO V8.0.3.0•Multi-objective•Discrete method•Multipoint crossover•Variable speed pumps•GA options
Key Points
Discrete method needs substantial storage space, especially for longer modeling periods and smaller
time intervals. Provides feasible solutions.
Adjusting parameters, such as modeling period, time intervals and hydraulic model details, are
important to obtain accurate results during reasonable running time.
Evolutionary algorithms are useful to optimize pumping.
Questions? [email protected]