Relevance of water demand forecasting increases with complexity of water supply systems. Several methods for water demand forecasting have been proposed in literature, mostly based on time-series analysis and machine learning, the later needing more detailed study of involved variables and choice of the best configuration to produce significant results. As an alternative to use of machine learning methods, this work presents two known methods of data approximation, namely, discrete Fourier series and Chebyshev polynomials, to real-time demand forecasting, through real-time updating of some adjustable coefficients. A real district of a water supply system is analyzed using these methods, showing a good approximation between measured and forecasted values. The most interesting point of application is agility of calculation and the fact that there is no need to have information on factors influencing water demand.