Forecasting model for real-time management: Application to wastewater networks

The work presented here aims at developing a flow forecast model dedicated to real-time management. The proposed model is based on the notion of a tra nsfer function for a linear system identified through the Kalman filter alg orithm. In a first step, the transfer function model is linked to the Muski ngum semi-empirical model; then it is modified to eliminate the autoregress ive component. The Kalman filter algorithm allows the parameters of the pro posed model to be updated upon the reception of each new measure with respe ct to the forecast errors observed in real time. To analyze the performance of the proposed model, its results are compared with those obtained using the dynamic wave model and the simplified kinematic wave model. Because of the absence of measured downstream flow values corresponding to the input h ydrograph, the results from the dynamic wave model are used as reference va lues to evaluate the performance of the other models. These results are als o used with the addition of noises to simulate measured values and feed, in "real-time," the identification algorithm of the transfer function in orde r to adjust, a posteriori, its parameters according to its differences in t he flow prediction. The results obtained by the transfer function model agr ee with those obtained by the dynamic model following the three performance criteria employed. The Nash coefficient and the ratio between the peak flo ws are close to unity in all of the cases. Also, the lag between the peak f lows estimated by the two models is negligible.