Leak detection and calibration of water distribution systems using transients and genetic algorithms

Abstract

The use of genetic algorithm optimisation applied to solving engineering problems has gained popularity over the last 10 years. Applications to the design of water distribution systems based on genetic algorithm optimisation first appeared in the early 1990s. This paper starts out with a brief review of the past use of genetic algorithms applied to aspects of water distribution systems. Leak detection and calibration of pipe internal roughnesses in a network are important issues for water authorities around the world. Computer simulation of water distribution systems has become a routine task of water authorities and consultants. One of the big unknowns in developing these models is the condition of the pipes, especially if they are old. It is very difficult to obtain reliable estimates of the roughness height for each pipe in the system using steady state calibration techniques. Liggett and Chen at Cornell University in 1994 developed an innovative technique called the inverse transient technique. The technique is able to determine, from unsteady pressure traces at a number of nodes in the network, the locations and magnitudes of any leaks that are occurring and the friction factor for each pipe in the network. An alternative approach to solving the minimization problem is presented in this paper. Genetic algorithm optimisation is used. A population of solutions is generated with each string representing values of the decision variables that are to be found. These include the magnitudes of leaks at nodes in the network and friction factors for each pipe. A forward transient analysis is performed for each string in the population that represents different combinations of leak magnitudes and friction factors. The sum of the absolute deviations between the measured transient pressures and the pressures predicted by the numerical model are determined and are used to determine the fitness of the string. The smaller the sum of the deviations then the larger the fitness that is assigned to the string. The genetic algorithm operators that are used include tournament selection, crossover and mutation. A new crossover operator is introduced. The genetic algorithm optimisation technique that has been developed in the research is applied to an example network. The results are encouraging and compare favorably with the inverse transient technique.