Regression-based inflow forecasting model using exponential smoothing time series and backpropagation methods for Angat Dam

Abstract

This paper deals with time series exponential smoothing and artificial neural network-based backpropagation methods in formulating a reservoir inflow forecasting model for Angat Dam in the Philippines. The predictive model is trained using dam daily average inflow observations inclusive of years 2003 to 2012, as recorded. Any real-time inflows forming a 5-consecutive-day vector could serve as input to the regression process. Its predictive power as measured by correlation coefficient at 0.959 for observed and predicted inflows taken from blind test set as well as 0.925 from validation set, could provide model users better perspective and outlook with regard to reservoir inflow conditions 24 hours into the future. In the background, this proposed model can offer dam managers protracted time in arriving at optimum reservoir water storage estimation as well as in load dispatching and scheduling when integrated into a decision support application.