Performance optimization of a leak detection scheme for water distribution networks

The subject of this paper is focused on the problem of robust leak detection in water distribution networks (WDNs). The main objective is to present the method for performance optimization of a model-based multipath leak detection scheme. The primary part of the robust fault detection scheme is realized applying model error modelling methodology. The model of the system is created by means of a neural network autoregressive model with an exogenous input, whilst the model error is identified using a linear autoregressive model with an exogenous input. The maximal performance of the method is achieved through an evolutionary optimization of the behavioural (relevant) parameters of the elementary blocks of the leak detection scheme. The merits and limitations of the method are discussed and highlighted taking into account experimental results obtained for leak detection in a real water distribution system.