Using Support Vector Machine Model for Fault Detection along a Water canal

Abstract

This paper reports a work in progress, the training of a Support Vector Machine model to detect faults in an experimental water supply canal. The work took place at the experimental canal of Núcleo de Hidráulica e Controlo de Canais at the Universidade de Évora. The main objective is to identify faults in the water depth sensors and to detect unauthorized water withdrawals using pattern recognition. The preliminary accuracy tests, in and out of sample, have shown an accuracy over 90% to identify 28 different patterns.