Le filtre de Kalman pour la prédiction multi-site des apports liquides annuels du bassin versant Cheliff.

Abstract

The objective the study is to investigate the applicability of the Kalman filter discrete to modeling and the multisite prediction of the contributions in the catchment area of high Cheliff (Algeria), as well as the improvement of the errors of prediction.This technique has the advantage of getting the error of prediction which presents an indicator of precision. Moreover, its algorithm works in the temporal field with a recursive nature, and has an optimal estimator in the direction of least squares. Another aspect of its optimality is that it incorporates all information available on the system, measurements and the errors, in an adaptive operator which is readjusted with each time a new measurement becomes available. For this purpose, the data of five (5) hydrometric stations in the catchment area of high Cheliff were invested over a period of 21 years (1990 to 2011) and then put into action. The results show that the multisite predictions more closely follow the recorded observations and the errors of prediction are minimal. This is valid as well in both dimensions, which shows that the KF is a rather powerful tool getting good predictions. In the end, the average error relative of the predictions is lower than 10% which is largely acceptable.