Predictive analysis of a water demand time series using a numerical method

The predictive analysis of a water demand time series is a complex and non-linear problem that cannot, simply, be solved using techniques from algebra, calculus, differential equations, and partial differential equations. Because of its intimate relationship with computers and computer architecture, a numerical method is proposed as a possible solution. The iteration direction, incremental or decremental, of the proposed technique is determined by the gradient between the first and the last element of the model´s input vector. In order to impose a fast rate of convergence to a solution, the iteration step size is biased by means of an offset. The results obtained demonstrate consistency on two different data sets, each containing a sum of 200 instances of data. Prediction accuracy values of 96.04% and 95.61% are obtained on the respective data sets, with rate of convergence values of 0.90 × 10⁶ i.s^-1 and 0.47 × 10⁶ i.s^-1.