The urban water supply is a huge problem faced in the modern world, once that it is frequently affected by leaks that leads to the waste of large amounts of water, time and money, three of the main pillars for the sustainable development of the cities nowadays. A quick detection of anomalous events such as leaks can bring significant gains to the companies. Such detection allows faster repairs of failures, ensuring a better attendance of the clients and smaller costs involving waste of water. Considering the needs of the development of an alarm system that acts readily when the system is under an anomalous event, this study aims to use computational tools of data analysis through neural networks for water demand predictions, and starting from the errors obtained from the comparison between the results of the predicted and the monitored data, detect the presence or not of anomalous events.