World Applied Sciences Journal 15 (3): 407-414, 2011 ISSN 1818-4952 © IDOSI Publications, 2011 Corresponding Author: Amir Jalalkamali, Department of Water Engineering, Faculty of Engineering, Islamic Azad University, Kerman Branch, Kerman, Iran. Tel: +98-913-343-1541. 407 Application of Hybrid Neural Modeling and Radial Basis Function Neural Network to Estimate Leakage Rate in Water Distribution Network Amir Jalalkamali and Navid Jalalkamali Department of Water Engineering, Faculty of Engineering, Islamic Azad University, Kerman Branch, Kerman, Iran Abstract: In this research the ability of a hybrid model of artificial neural network (ANN), feed forward networks (FFN) and recurrent neural networks (RNN), is investigated with genetic algorithm (GA). GA is used in order to determine the optimal structure of ANN (i.e. the number of neurons for each hidden layer). Furthermore, hybrid model's results are compared with radial basis function neural network (RBF). A water supply network located in Kerman, Iran is considered as case study in order to illustrate the efficiency of the modeling procedure. Obtained results apparently show that the ANN-GA models can be used successfully to estimate leakage rate in water distribution networks. In addition, a comparative study of models indicates that the feed forward networks hybrid with GA performed better than the other models. Key words:Leakage Water distribution network Feed forward network Recurrent neural network Radial basis function Genetic algorithm INTRODUCTION control setting of pumps and valves up to 24h rolling Leakage is one of the most serious problems in demand. Salomons et al. [2] described the Haifa-A municipal water distribution networks. Limited resources hydraulic network, the ANN predictor, the GA optimizer of water special in arid area and the increasing expenses and the demand forecasting model that were used. of transport, treatment, pumping, storage and distribution Mounce et al. [3] presented the online application of of water, notify the importance of leakage reduction. artificial neural network and fuzzy inference systems for Also water quality problems could result from pollution at detection of leakage in real water distribution system. leak points. Due to the direct relation between leakage and Nazif et al. [4] used a genetic algorithm based optimization pressure, pressure monitoring is a useful and cost model to develop the optimal hourly water level variations effective method for leakage reduction. On the other hand in a storage tank for minimizing the leakage level in the relationship between pressure changes and the rate of different seasons. Koppel et al. [5] showed that leakage in water supply networks is non-linear and the Levenberg-Marquardt algorithm may successfully be complex. Mathematical hydraulic models can be utilized used for calibration of water distribution system model. for modeling the leakage rate in terms of pressure Rao et al. [6] described that the ANN is employed in changes. The problems in developing these models are preference to the hydraulic simulation model within the twofold: firstly, understanding of complex concepts optimization process. Bowden et al. [7] developed general related to the field of hydraulic, secondly, wasting a regression neural networks (GRNNs) for forecasted plenty of time for driving a suitable mathematical relation. chlorine residuals in the Myponga water distribution In recent years, successful application of soft system. Mounce et al. [8] presents the application of computing techniques in water distribution networks have artificial neural network (ANNs) for analysis of data been widely published, in the following a short review of from sensors measuring hydraulic parameters these studies is given. Martinez et al. [1] used an artificial (flow and pressure) of the water flow in treated water neural network (ANN) predictor in place of the EPANET distribution systems. Celia et al. [9] used ANN models model and a dynamic genetic algorithm to optimize the to predict residual chlorine, substrate and biomass operating horizon, in response to a highly variable