Performance assessment among hybrid algorithms in tuning SVR parameters to predict pipe failure rates

Abstract

Pipe failures often occur in water distribution networks and result in large water loss and social-economic damage. To reduce the water loss and maintain the conveyance capability of a pipe network, pipes that experienced a severe failure history are often necessary to be replaced. Several studies and methods have been introduced for predicting failure rates in urban water distribution network pipes by researchers, each of them has some special features regarding the effective parameters and many methods such as Classical and Intelligent methods are used, leading to some improvements. In this paper, the method incorporates hybrid support vector machine and heuristic algorithms techniques for efficient tuning of SVM meta-parameters for predicting water distribution network. Performance results are Compared with continuous genetic algorithm-based SVR (SVR-GA), continuous ant colony algorithm-based SVR (SVR-ACO), particle swarm optimization-based SVR (SVR-PSO), artificial neural networks (ANNs) and adaptive neuro-fuzzy inference systems (ANFIS).