Determination of Effective Parameters in Pipe Failure Rate in Water Distribution System Using the Combination of Artificial Neural Networks and Genetic Algorithm

In water supply systems, the accidents occurring in pipes are of the utmost importance and sensitivity. Failure of the pipes is not necessary with the end of their life and different factors namely age, diameter, material, stability and corrosion of soil and water, execution, installation and operational conditions such as hydraulic pressure are effective on it. At the same time studies show non comprehensiveness of presented relations in prediction of pipes failure rate. In this research, with regards to the available software and hardware, a structure was developed using combination of new optimization and simulation models. In this structure, theartificial neural networks were used for simulation of the pipes failure rate. Considering the point that neural networks are always consider as a black box and unable to provide the effect of each independent variable on the dependent variable, on the other hand, they are prone to incorrect training. Therefore, in this study to determine the input parameters affecting failure rate and the most appropriate structure of artificial neural network (as well as bios vectors, layers adjusting weights and the number of neurons), Genetic algorithm has been used with the aim of presentation of a structure which has the minimum error rate of simulation.In this algorithm, decision variables and properties of neural networks are the parameters affecting the failure rate. By running the developed optimization model, in addition to the effective parameters on pipe failure, the best neural network structure for simulation pipe failure rate can be determined. The advantage of the proposed method is full coordination between the input parameters and network structure in prediction of the pipes failure rate. The results of this study can be used to find the most appropriate relationship failure rate pipes with regards to the effective parameters and take necessary actions for decision making lead to resolve problems due to it. The results of this research indicating that the proposed combination method is able to extract the optimal and effective parameters on the pipes failure rate amongst the factors affecting failure rate, and also have been caused improve power capabilities and expansion of neural network structure that indicate high efficiency of the proposed method in simulation of nonlinear and complex relations.