Continuing Lines Heuristic optimisation techniques for spatial environmental problems with multiple

1
Continuing LinesHeuristic optimisation
techniques for spatial environmental problems
with multiple objectives

• Kees Vink

2
Program

• Introduction
• Applications
• model calibration
• regional drinking watyer supply
• Conclusions
• Discussion

3
Urbanisation is an ongoing process
4
An increasing number of drinking water production
sites pumps urban groundwater
5
Urban areas complex environments

• spatial and temporal complexity
• many stakeholders

6
Urban watermanagement and drinking water supply

• problems and opportunities
• conflicting interests
• nonlineair relations
• need for apropriate techniques

7
Case 1Multiple objective calibration of a
groundwater model by means of a genetic algorithm

• Kees Vink

8
Groundwater model parameter optimisation

• urban environment spatial and temporal
  complexity
• parameter value uncertainty
• combinatorial explosiveness
• identification problem

9
Calibration as an optimisation problem with two
conflicting objectives

• Minimum differences between simulated and
  observed heads
• Minimum deviation from initial parameter value
  estimates

10
Objective function 2Minimal deviation from
initial
11
Reasons for using more than one objective

• A model is imperfect (not only parameter
  estimation errors but also conceptual errors,
  reference data errors). Minimising the
  differences between simulated and observed heads
  only may result in a sub-optimal calibration
  adjusting the parameter configuration for
  compensation of conceptual and reference data
  errors
• With only one objective function equifinality is
  more of a problem than when using multiple
  objective functions

12
Conflicting objectives Pareto front
Impact 2
Impact 1
13
Genetic algorithms

• Population of solutions
• Fitness
• Crossover
• Survival and reproduction of the fittest
• Suitable for many complex optimisation problems
• Suitable for multiple objective opt. problems
• Heuristic techniquegt flexible

14
Genetic algorithms Principles of Darwins
evolution theory
Natural selection Variation Survival of the
fittest
15
Optimisation with a genetic algorithm
16
How do we define fitness?

• Pareto efficiency !
• Uniqueness of a solution

demo
17
Results
Initial
Final
18
Case 2Multi-objective Optimisation of Regional
Drinking Water Supply with a Genetic Algorithm

• Kees Vink
• Paul Schot
• Utrecht University
• Netherlands

19
Drinking Water SupplyConsumers
20
Drinking Water SupplyProduction Wells
21
Drinking Water SupplyTransport Network
22
Production Strategies Required Due to

• Changes of conditions
• Spare capacity

23
Production StrategyHow to Distribute Production
Rates Over

• Production wells
• Purification plants
• Transport system

24
Many Strategies Are FeasibleEach With Different
Impacts
25
Strategy Definition
Traditional Approach

• Expert judgment
• Intuition

Contemporary Approach

• Decision Support System

26
Evaluation of Strategies byQuantifying and
Comparing
27
Quantification of Impacts
28
Multi-objective Optimisation requires Valuation

• Explicit / Implicit
• A Priori / A Posteriori

29
Pareto Front
30
Pareto Front
31
Decision Support
32
Quality control validation?

• Heuristic techniques are tricky
• Try to verify results by deduction
  circumstantial validation
• Single objective extreme solutions
• Insensitive parameters
• dummy model tests

33
End of PresentationThank you for your
attention