DescriptionSizeFormat
01front.pdf80.2 kBAdobe Acrobat PDF View/Open
02whole.pdf1.8 MBAdobe Acrobat PDF View/Open

Please use this identifier to cite or link to this item: http://hdl.handle.net/1959.13/24965

Title
Optimisation of Urban Water Supply Headworks Systems Using Probabilistic Search Methods and Parallel Computing
Author/Creator
Cui, Lijie
Institution
University of Newcastle. School of Engineering (Civil).
Description
Realistic optimisation of the operation and planning of urban water supply headworks systems requires that the issues of complexity and stochastic forcing be addressed. The only reliable way of accomplishing this is to use simulation models in conjunction with the Monte Carlo method which generates multiple hydro-climate replicates. However, such models do not easily interface with traditional optimisation methods. Probabilistic search methods such as the genetic algorithm (GA) and the shuffled complex evolution method (SCE) can be coupled to a generalised simulation model and thus accommodate complexity as well as stochastic inputs. However, optimisation of complex urban water supply systems is computationally intractable if Monte Carlo methods have to be used. This study first compared the GA and the SCE method using a simple case study. Both methods were found to cope well with the piecewise flat objective function surface typical of the headworks optimisation problem. This is because they have the inherent capability of vigorously exploring beyond the domain of a flat region. The SCE method is recommended especially when fast location of a good solution is desired. Nonetheless, the GA was preferred due to its inherent parallelism. Two methods were then explored to improve computational efficiency and turnaround time: parallel computing and replicate compression. The Sydney headworks system was used as a case study to investigate the key aspects of a full-scale headworks optimisation. It was concluded that the speedup was nearly proportional to the number of processors employed. Replicate compression can very significantly reduce the computational turnaround time for Monte Carlo simulation; unfortunately, this conclusion must be tempered by the limitation that the objective function depends on penalties arising from restrictions only. Critical analysis of the GA results suggested the optimised results were sound. The case study demonstrated the feasibility of parallel GA to identify near-optimal solutions for a complex system subject to stochastic forcing.
Description
PhD Doctorate
Date
2003
Keyword
parallel computing
Keyword
Optimisation
Keyword
Water Supply
Keyword
genetic algorithms
Keyword
Monte Carlo
Keyword
simulation models
Keyword
water resources
Resource Type
Australasian Digital Thesis
Rights
http://www.newcastle.edu.au/copyright.html
Rights
Copyright March 2003 Lijie Cui
Identifier
http://hdl.handle.net/1959.13/24965
Identifier
uon:685
Language
eng
Full Text
Full Text
 
Image Thumbnail