Author(s): Lijie Cui; George Kuczera
Linked Author(s):
Keywords: Parallel computing; Genetic algorithm; Replicate compression; Optimization
Abstract: At present optimization of complex multi-reservoir urban water supply systems is computationally intractable if Monte Carlo methods are used to simulate stochastic hydro-climatic inputs. Two promising approaches are examined for improving the computational efficiency of such optimization problems. The first is to implement a genetic algorithm (GA) based optimization model in a parallel computing environment. The performance of parallel GA is assessed using a simple case study involving one reservoir and one urban demand with three decision variables. The case study shows that elapsed time can be reduced with a speedup factor of up to 3.5 using four processors. Further improvement can be achieved using replicate compression. This compression scheme simulates a lumped model of the system to identify critical periods. The full simulation model is then applied to each critical period. For high reliability urban systems this scheme can very significantly reduce the Monte Carlo simulation effort for multi-reservoir systems.
Year: 2001