ROUTINE FORECASTING OF THE DAILY PROFILES OF HOURLY WATER DISTRIBUTION IN CITIES. AN EFFECTIVENESS ANALYSIS

摘要

Sample results have been presented of verifying three groups of methods of forecasting the time series of short-duration water distributions in city water grids. The analysis covered: ARIMA class models, the time series exponential smoothing methods and artificial neural networks. Since chronological sequences of observations from the immediate past were analyzed, the adopted models did not take any external variables into account. The forecasting errors in the case of multilayer perceptron neural networks were found to be comparable or smaller than the errors of prediction by the ARIMA class models and by the methods of the exponential smoothing of time series.