Recent advances in our flow estimation model
During the first months of 2017, ACCIONA Agua and Homeria have continued researching and improving a complex algorithm to detect leakages of water in Burgos by using a pattern based approach. From the original time series, input and output patterns are constructed in such a way that weekly and annual cycles are filtered. The input pattern is used in a k-nearest algorithm in order to obtain a training set. The estimation of the water flow for a particular day is obtained as the average of the training set output patterns.
As an illustration of the algorithm, the following figures show the estimation for March 24th, 2015 in Villalonquejar DMA. The first figure depicts the X-pattern for such date (black solid line) and the corresponding 5 nearest neighbours (color dots) while the second figure shows the corresponding forecast for March 25th, 2015 (solid red line) together with the actual measured data (black dots).