MODELISATION ET ANALYSE DE LA SERIE CHRONOLOGIQUE DE PRODUCTION D’EAU DE CONSOMMATION PAR LISSAGE EXPONENTIEL

Abstract

The time series analysis and forecasting especially the short and medium term, has been an important development. Forecasting is fundamental to the extent that it is based on the optimal decision making. Predict the future behavior of a time series requires the use of several methods of forecasting, because we found the reliability of a prediction method depends not only on its theoretical complexity, but also data, the available information and application fields. The objective of modeling the observed time series is to predict its future behavior by the determination of an appropriate smoothing of the observed series data. In this study the problem of smoothing is treated using exponential smoothing which allows giving more weight to recent values of the time series. A predictive model for the production of drinking water from fields capturing of Biskra city is proposed. Weekly data from the Public authority (ADE) cover the period from January 2009 to December 2011. The results led to the forecast horizon given after calibration and validation model and provide some answers in the form of long-term series and residual series obtained (observed series - smoothed series).